# Midas Knowledge Base

> The operating knowledge of MIDAS (AI automation agency) plus applied playbooks for AI in ecommerce and building with coding agents. Three interlinked knowledge bases, 436 markdown pages.

This file is an index for AI agents (llmstxt.org format). Each link below points to a raw markdown page. Read this map first, then fetch the specific pages you need. To load an entire knowledge base in one request, fetch its full bundle under `/kb/_full/<name>.txt`. The complete corpus is at `/kb/_full/all.txt`.

## AI Agency Playbook

How to run an AI automation agency end to end: strategy, lead gen, cold email, sales, fulfillment, scaling, plus 60+ buildable system blueprints. Built from the Nick Saraev / Maker School curriculum and real MIDAS campaign data.

Full bundle: [ai-agency.txt](https://midas-wiki.vercel.app/kb/_full/ai-agency.txt)

### Blueprints
- [Ad Creative Spinner Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/ad-creative-spinner.md): Turn winning static image ads into hundreds of branded variations using AI image editing. A creative agency drops source ads into a folder, chats with an AI agent, and gets spun variants in an output folder. Target buyer: PPC agencies, c...
- [Agency Website Structure — What Belongs Above the Fold](https://midas-wiki.vercel.app/wiki/blueprints/agency-website-structure.md): Top-to-bottom structure for an agency landing page. Tested by Nick across Maker School / LeftClick / 1SecondCopy. Services section is the LEAST important part.
- [Agent To-Do Queue Dispatch Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/agent-todo-queue-dispatch.md): An agentic personal operating system built by repurposing Linear (normally a software issue tracker) as a to-do board that feeds tasks to always-on Claude agents. You add cards, tag the delegatable ones, and agents automatically pick the...
- [AI Graphic Design Agent (N8N Live Build)](https://midas-wiki.vercel.app/wiki/blueprints/ai-graphic-design-agent.md): A chat-based AI graphic design agent built in N8N. Users interact via embedded chat widget, agent generates logos, style guides, gradient backgrounds, and edits/revises images. Sellable for $2,000+ per installation.
- [AI-Native Business Setup — Solopreneur Stack](https://midas-wiki.vercel.app/wiki/blueprints/ai-native-business-setup.md): End-to-end setup for running an AI-native solopreneur business. Knowledge base + Claude Code as command center + Modal for deployment + UniPile for social. From Miguel Torrez.
- [App Store Submission Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/app-store-submission.md): End-to-end submission flow for shipping a Claude-Code-built Expo app to the iOS App Store + Google Play Store. Source: Nick Sarayev's 2026 mobile course.
- [Auto Invoice Collection System Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/auto-invoice-collection-system.md): An automated dunning system that eliminates manual follow-up on outstanding invoices — a major pain in the automation-agency space where clients sit on net-30 terms. It queries your invoicing platform once a day, calculates how long each...
- [Auto Research Dashboard for Website Optimization](https://midas-wiki.vercel.app/wiki/blueprints/auto-research-dashboard.md): Live dashboard tracking autonomous performance optimization experiments.
- [Automated Hiring Pipeline Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/automated-hiring-pipeline.md): A ClickUp + make.com recruitment funnel that filters candidates at scale, gates every hire behind a paid trial, and turns pipeline status changes into automated candidate communication. Built for agencies growing past a freelance outfit...
- [Blog Article Generator Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/blog-article-generator.md): AI-powered blog article engine: web research → mega outline → per-section writing + illustrations → Google Doc. Sell price: ~$1,000 per deployment; also packaged as part of the B2B agency niche pack at $1.7K+ and sellable to any content,...
- [Bookkeeping Organizer Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/bookkeeping-organizer.md): A "senior bookkeeper" AI prompt/skill that turns raw bank statements and transaction exports into clean, categorized, tax-ready records. The accountant (or business owner) supplies high-level category instructions; the agent does the lin...
- [Browser Automation Pattern (Chrome DevTools MCP)](https://midas-wiki.vercel.app/wiki/blueprints/browser-automation-pattern.md): A reusable pattern for automating any browser-based task using Chrome DevTools MCP.
- [Claude Skills Productized Package](https://midas-wiki.vercel.app/wiki/blueprints/claude-skills-productized-package.md): Pattern for selling Claude skills as mini products to non-technical clients. Different from agency work — closer to "productized SaaS" without writing real SaaS.
- [Client Offer Ideation Process](https://midas-wiki.vercel.app/wiki/blueprints/client-offer-ideation-process.md): Live SOP for building a client's cold email offer from a kickoff call. Demonstrated in Call 42. Produces a deployable offer in under 30 minutes.
- [Client Portal Builder — Full-Stack Deliverable](https://midas-wiki.vercel.app/wiki/blueprints/client-portal-builder.md): Build a complete client portal with onboarding doc + weekly reports + interactive dashboard + RAG chatbot. ~30 min for MVP via Claude Code. Premium retainer differentiator.
- [Coaching CRM Pipeline (ClickUp) Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/coaching-crm-pipeline.md): A client-management CRM built on top of ClickUp for coaches running a hands-on program. It gives the coach one standardized place to pump every client through the program, from intake to completion, with per-client asset storage and two...
- [Cold Email Campaign Generator Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/cold-email-campaign-generator.md): Generates complete cold email campaigns with A/B testing variants and follow-up sequences, based on a client's niche and offer, using high-performing existing campaigns as templates.
- [Cold Email Engine Hosted for Clients](https://midas-wiki.vercel.app/wiki/blueprints/cold-email-hosting-for-clients.md): Deliver the entire cold email machine as a client-owned system. Three implementation paths depending on complexity needs.
- [Cold Email Optimizer (Autoresearch Implementation)](https://midas-wiki.vercel.app/wiki/blueprints/cold-email-optimizer-autoresearch.md): Live implementation of Karpathy's auto research loop applied to cold email optimization.
- [Cold Email Reply Classifier — Automated Reply Handler](https://midas-wiki.vercel.app/wiki/blueprints/cold-email-reply-classifier.md): Auto-classify and reply to positive cold email responses at scale. 80% as good as human-written. Acceptable for 5K+ emails/day volume.
- [Content QA Feedback Loop — Human + AI Hybrid](https://midas-wiki.vercel.app/wiki/blueprints/content-qa-feedback-loop.md): Content pipeline with mandatory human QA + structured rejection logging. Designed during Mr. Beast session (Call 48). Pattern: AI for production, humans for QA. Never the reverse.
- [Content Repurposer Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/content-repurposer.md): Takes a transcript and produces multiple content formats in parallel using sub-agents.
- [CRM Follow-Up Nurture System Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/crm-follow-up-system.md): Automated, personalized follow-up emails to every lead in your CRM pipeline, sent in existing email threads, matching the tone of previous conversations.
- [Directive Template](https://midas-wiki.vercel.app/wiki/blueprints/directive-template.md): The exact structure for writing directives in the DOE framework.
- [Facebook Ads Spy Tool Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/facebook-ads-spy-tool.md): Scrape the Facebook Ad Library, categorize ads by type (video/image/text), analyze each with appropriate AI model, generate summaries + rewritten ad copy + image/video prompts. Sell price: $2,000+. Build time: ~2 hours.
- [Free Trial Close — 30-Minute Build → $15-20K LTV](https://midas-wiki.vercel.app/wiki/blueprints/free-trial-close-technique.md): When a high-value prospect wants proof, build the demo live in 30 minutes for free. The math wins.
- [Full Fulfillment Pipeline Blueprint (Leftclick)](https://midas-wiki.vercel.app/wiki/blueprints/full-fulfillment-pipeline.md): Complete end-to-end pipeline for a B2B cold email agency — from sales call to automated replies.
- [Hiring-Signal (Search-Intent) Outreach Scraper Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/hiring-signal-outreach.md): A "search intent scraping" system: it treats a company posting a job for a role as a buying signal, scrapes those hiring companies, enriches the decision maker, writes a personalized first line, and drops the lead into a cold-email campa...
- [Image Gen Visual Reasoning Pipeline](https://midas-wiki.vercel.app/wiki/blueprints/image-gen-visual-reasoning-pipeline.md): Use 2026 image models (Nano Banana Pro, GPT Image Gen 2) for business use cases beyond "cute images" — visual reasoning, infographics, before/after, audience-aware diagrams. Agency monetization included.
- [Inbox Cleaner Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/inbox-cleaner.md): AI-powered email triage — reads all unread emails, classifies importance, marks non-important as read.
- [Instagram Parasite System Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/instagram-parasite-system.md): Scrape competitor Reels daily, transcribe via Whisper, AI-filter for relevant topics, research via Perplexity, rewrite as new scripts. Nick used this to go from 0 to 10,000 followers in 15 days. Sell price: $3,000-5,000. Build time: ~1 h...
- [Instantly Campaign Operations Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/instantly-campaign-operations.md): Complete operational setup for running cold email campaigns in Instantly, from naming conventions to reply handling.
- [Invoice Data Extractor Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/invoice-data-extractor.md): Takes PDF invoices and returns structured JSON with all financial data extracted.
- [KPI Reporting Automation Without Native API](https://midas-wiki.vercel.app/wiki/blueprints/kpi-automation-no-api.md): Build KPI dashboards for platforms that DON'T have direct APIs by using Zapier integrations as a de facto API. Live demoed for Paint Scout in Call 35.
- [Lead Enrichment + HubSpot Pipeline](https://midas-wiki.vercel.app/wiki/blueprints/lead-enrichment-hubspot-pipeline.md): End-to-end enterprise lead enrichment system with dedup + AI personalization. Built by Rohan for an enterprise client (Call 48). Sells for $1,600 MVP → $10K+ full system.
- [Lead Scraping Pipeline Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/lead-scraping-pipeline.md): End-to-end lead generation: scrape → verify → scale → classify → enrich → upload.
- [LinkedIn Connection-Request Automation (Phantom Buster) Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/linkedin-connection-request-automation.md): A system that auto-sends LinkedIn connection requests with AI-customized icebreakers to an audience defined in plain English. The operator types who they want to reach (e.g. "digital marketing agencies in the US with 1-10 employees"); th...
- [LinkedIn Parasite System Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/linkedin-parasite-system.md): Fully autonomous LinkedIn content engine that scrapes viral posts from top creators, researches and rewrites them with unique twists in your tone of voice, and auto-posts on schedule. Sell price: $1,500+ per deployment.
- [LinkedIn Sales Navigator Lead Scraper Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/linkedin-sales-navigator-scraper.md): Natural language lead scraping from LinkedIn Sales Navigator — the highest quality source for B2B lead data.
- [Meeting Notes Processor Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/meeting-notes-processor.md): A pure SOP skill (no scripts required) that converts meeting transcripts into structured action items.
- [Mobile App Pipeline Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/mobile-app-pipeline.md): End-to-end Claude Code pipeline for shipping a real mobile app (iOS + Android) from zero to App Store submission. Source: Nick Sarayev's 4-hour mobile app course. Three example apps built in the course: habit tracker, Cal AI lookalike (c...
- [Multi-Chrome Parallelization Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/multi-chrome-parallelization.md): Scale browser automation from 33 hours to 20 minutes using parallel agent instances.
- [n8n → Agentic Workflow Conversion](https://midas-wiki.vercel.app/wiki/blueprints/n8n-to-agentic-conversion.md): Convert existing n8n / Make.com flows to agentic workflows (Claude Code, Modal, trigger.dev) without rebuilding from scratch. Live demoed in Call 34.
- [Parallel Website Designer Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/parallel-website-designer.md): Generate high-quality websites using screenshot-based design reference + verification loops + parallel agent exploration.
- [PR / Organic-Media Authority System (SOS + Qwoted + Auto-Responder) Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/pr-authority-sos-qwoted.md): A system for manufacturing authority and credibility through organic (earned) media placements — getting quoted by journalists as an "AI/automation expert" so you can command higher pricing. It combines two free journalist-query services...
- [Full-Stack Proposal App Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/proposal-app.md): A PandaDoc-alternative built in ~15-20 minutes with plan mode.
- [Reddit Organic-Seeding Campaign Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/reddit-organic-seeding-campaign.md): A system for putting more eyeballs on a brand by softly recommending its product inside relevant Reddit threads at scale. AI monitors subreddits and pre-drafts human-sendable recommendation snippets; humans post them across a variety of...
- [Sales Call Intelligence Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/sales-call-intelligence.md): A transcript-mining system for agencies and sales teams. Feed it one sales-call transcript or a whole stack, and it extracts the recurring signal buried in your calls, then turns that signal into a sharper script, an objection-handling g...
- [Agency Sales Metrics Dashboard Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/sales-metrics-dashboard.md): A self-updating dashboard for an AI agency that tracks exactly four sales metrics — meetings booked, proposals sent, deals closed, and revenue defined as cash collected — by piping each business event into a Google Sheet and visualizing...
- [Security Audit Flow Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/security-audit-flow.md): Multi-agent security audit pattern using fresh, unbiased agent instances.
- [SEO + AI-Visibility (AEO) Audit Loop Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/seo-aeo-audit-loop.md): An audit system that scores a website against standardized benchmarks for BOTH classic search engines and AI-answer engines (AEO — Answer Engine Optimization), then iteratively closes the biggest gaps until every priority query maps to a...
- [Signal Forge — CRM Intent Signal Enrichment](https://midas-wiki.vercel.app/wiki/blueprints/signal-forge-crm-enrichment.md): Weekly automated research on sleeping CRM leads. Detects intent signal changes → surfaces warming leads automatically. Built for long-sales-cycle B2B (manufacturing, logistics, enterprise SaaS).
- [Google Slides Proposal Generator Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/slides-proposal-generator.md): Form-triggered pipeline: sales call notes → LLM generates structured proposal content → copies Google Slides template → replaces placeholders → emails client. Internal agency tool.
- [Speed-to-Lead Call System Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/speed-to-lead-call-system.md): A phone-follow-up system that maximizes pickup rate on opted-in ad leads. Built for any business running ad funnels with phone follow-up — dentists, clinics, and other high-volume call funnels doing several thousand calls/day. This becam...
- [System Prompt Template (agents.md / claude.md)](https://midas-wiki.vercel.app/wiki/blueprints/system-prompt-template.md): The three things every system prompt must include for effective agentic work.
- [Upwork Flowchart Pitch Generator Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/upwork-flowchart-pitch-generator.md): A system that turns a raw Upwork job description into a polished-looking system flowchart in ~5 seconds, so you can record a Loom walking through "your" proposed architecture and apply for automation jobs without first scoping the system...
- [Upwork Job Scraper Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/upwork-job-scraper.md): Scrape Upwork jobs, filter serious buyers, generate customized proposals, and output ready-to-apply package.
- [Vibe-Coded SaaS App Blueprints](https://midas-wiki.vercel.app/wiki/blueprints/vibe-coded-saas-apps.md): Five complete apps built live in the Vibe Coding course, from simple portfolio to full SaaS with payments.
- [Video-to-Action Pipeline Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/video-to-action-pipeline.md): Teach agents from video tutorials — extract steps via Gemini's video understanding, execute via Claude.
- [Voice-AI Agent Service Business Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/voice-ai-agent-service.md): A productized service that builds inbound voice AI agents (phone answering, booking, intake) for local service businesses whose owner-operators answer their own phones. Corey (My Sick Builds) runs this model; pricing is $5-25K per projec...
- [Voice-Matching Perplexity Humanizer Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/voice-matching-perplexity-humanizer.md): A paste-into-Claude protocol Nick calls "the Nick humanization protocol." It rewrites AI-generated text sentence by sentence so it reads in your own voice — Nick's goal is to move typical AI output from ~70% "sounds like you" (his estima...
- [Website Builder for Outreach Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/website-builder-outreach.md): Generate high-quality, unique websites for cold outreach prospects as a free value-add — the ultimate knowledge arbitrage play.
- [YouTube Comment Idea Generator Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/youtube-comment-idea-generator.md): A system that turns your own video's YouTube comments into a content calendar of new video ideas. Built for coaches, consultants, and any YouTube-inbound business, and sold at $1.5K+. [Source: Maker School — Resource Library 5. Niche Packs]
- [YouTube Trend Detector Blueprint](https://midas-wiki.vercel.app/wiki/blueprints/youtube-trend-detector.md): Two-workflow system that monitors YouTube channels, tracks video performance over time, and sends daily email digests highlighting outlier videos by view-count multiples. Inspired by 1of10.co. Sell price: $2,000+ per build. Build time: ~...
### Claude Code
- [Always-on Claude Code Home Server (Mac-Mini Style)](https://midas-wiki.vercel.app/wiki/claude-code/always-on-home-server.md): Run an old MacBook, hardwired and left on 24/7, "Mac Mini style" as a persistent back-end server running Claude Code agents constantly. With Chrome DevTools MCP connectors wired in, the agent can browse the internet, and you can jump int...
- [CLAUDE.md — The Always-On System Prompt](https://midas-wiki.vercel.app/wiki/claude-code/claude-md-system-prompt.md): CLAUDE.md (must be capitalized C-L-A-U-D-E, lowercase .md) in your root directory is automatically prepended to every new conversation as a system prompt.
- [Context Management](https://midas-wiki.vercel.app/wiki/claude-code/context-management.md): Context window = fixed budget (200K for Opus 4.6). Every token consumed reduces available capacity AND output quality.
- [Permission Modes](https://midas-wiki.vercel.app/wiki/claude-code/permission-modes.md): Claude Code has 4 main permission modes plus extras. Cycle between them with Shift+Tab in terminal or click the mode indicator in GUI.
- [Plan Mode](https://midas-wiki.vercel.app/wiki/claude-code/plan-mode.md): When building something even slightly complicated, Claude Code automatically switches to plan mode (indicated by a blue color in the bottom-left corner, replacing the red bypass-permissions indicator).
- [Progressive Disclosure — Context Window Optimization](https://midas-wiki.vercel.app/wiki/claude-code/progressive-disclosure.md): Only load information into the model's context window when it's actually needed.
- [Prompt Caching Economics](https://midas-wiki.vercel.app/wiki/claude-code/prompt-caching.md): Prompt caching lets an LLM provider resume from a prefix of your prompt instead of reprocessing the whole context on every request. Within the cache window you are billed only for the new tokens you add on top of the cached prefix, not t...
- [Self-Annealing — Skills That Heal Themselves](https://midas-wiki.vercel.app/wiki/claude-code/self-annealing.md): The defining power of skills: they are self-annealing over time. When the agent runs a skill and encounters an error (API rate limit, script bug, missing knowledge, service outage), it:
- [Skills Architecture](https://midas-wiki.vercel.app/wiki/claude-code/skills-architecture.md): Skills are the evolution of standard operating procedures (SOPs), translated for AI agents instead of humans. Where SOPs standardize work for human employees, Skills standardize work for AI agents. The key difference is translating instr...
- [Token Conservation](https://midas-wiki.vercel.app/wiki/claude-code/token-conservation.md): Token length scales inversely with output quality — more tokens in context = more money AND lower quality. Conservation is both an economic and a quality imperative.
### Frameworks
- [The 60/30/10 Rule for Model Selection](https://midas-wiki.vercel.app/wiki/frameworks/60-30-10-model-selection.md): Use a top-level router (smartest model) to assign tasks to appropriately-sized models. Saves ~60% on token costs.
- [Accountability Groups for Execution Consistency](https://midas-wiki.vercel.app/wiki/frameworks/accountability-groups.md): A small group of people doing the same thing as you, run on a required daily check-in, to counteract the week-one motivation drop and keep you executing over long periods.
- [AI Spend Scaling Heuristic](https://midas-wiki.vercel.app/wiki/frameworks/ai-spend-scaling-heuristic.md): When AI subscription/token cost is under 10% of the revenue it drives, stop economizing and scale the plan up to remove even minor friction.
- [App Design 5-Step Framework](https://midas-wiki.vercel.app/wiki/frameworks/app-design-5-step.md): Nick's pre-build ideation framework for mobile (or any) apps. Defines MVP scope before any code. Skipping this = bloated apps with too many screens and no retention.
- [Auto Research Framework (Karpathy's Method)](https://midas-wiki.vercel.app/wiki/frameworks/auto-research.md): A self-improving optimization loop that runs autonomously. Set the goal, define the metric, let the agent iterate. From Karpathy's github.com/karpathy/autoresearch.
- [Build-Test-Iterate Loop](https://midas-wiki.vercel.app/wiki/frameworks/build-test-iterate-loop.md): The standard workflow for creating and refining any skill.
- [Business-Model Revenue Curves: Agency vs SaaS vs Info Product](https://midas-wiki.vercel.app/wiki/frameworks/business-model-revenue-curves.md): Three business archetypes plotted as revenue-over-time have different shapes, and the shape dictates what you should start when you're broke. Agency spikes fast then plateaus; SaaS is flat for a long time then goes exponential; an info p...
- [Buyer-Awareness Ladder](https://midas-wiki.vercel.app/wiki/frameworks/buyer-awareness-ladder.md): A three-rung model that ranks prospects by distance-to-close, so you know how many separate "sells" a lead requires before they buy. Also called buyer awareness, buying temperature, or the buyer ladder.
- [Claude as a Contractor (OCD Work-Order Framework)](https://midas-wiki.vercel.app/wiki/frameworks/claude-as-contractor.md): Treat Claude like a contractor, not a genie or a friend you text — brief it with a full scope of work using the OCD structure (Outcome, Context, Constraints, Definition of Done), front-loaded in one message.
- [Co-founder Selection: You Become the Average of Your Partners](https://midas-wiki.vercel.app/wiki/frameworks/co-founder-selection.md): Choose co-founders whose strengths cover your gaps, because their pull reshapes you: you end up being the sum total, or the average, of everyone you build with.
- [Compound Probability Problem](https://midas-wiki.vercel.app/wiki/frameworks/compound-probability.md): If Claude is independently 90% successful on a task:
- [DOE Framework (Directive-Orchestration-Execution) — OUTDATED](https://midas-wiki.vercel.app/wiki/frameworks/doe-framework.md)
- [Effectiveness over Efficiency](https://midas-wiki.vercel.app/wiki/frameworks/effectiveness-over-efficiency.md): Do the thing manually at least the first 100 times before you optimize it. Speed and output quality beat process efficiency in a fast-moving market.
- [Options EV Ranker](https://midas-wiki.vercel.app/wiki/frameworks/ev-ranking.md): A prompt-driven decision framework that turns "what should I do?" into a ranked menu of options scored by expected value, instead of returning one generic answer you can't act on.
- [Friction & the Lifestyle Audit](https://midas-wiki.vercel.app/wiki/frameworks/friction-lifestyle-audit.md): A method for auditing your life for "friction" — anything that makes it harder to do what you want to do — by listing 50-100 friction points across six prompt sets, then ranking them by difficulty and solving easiest-first.
- [The Factoring Method (Consultant Goal Decomposition)](https://midas-wiki.vercel.app/wiki/frameworks/goal-factoring.md): A consultant strategy method: put a goal at the top, then recursively break it into sub-levers until you reach atomic, executable actions at the bottom. The output is a concrete deliverable list you dump into your PM tool and execute top...
- [The Iceberg Technique (Strategic Context Loading)](https://midas-wiki.vercel.app/wiki/frameworks/iceberg-technique.md): Only 10-20% of available information should be in the context window at any time. The rest stays accessible on-demand.
- [Knowledge Arbitrage Framework](https://midas-wiki.vercel.app/wiki/frameworks/knowledge-arbitrage.md): You have access to AI tools that your prospects and clients don't. Leverage this asymmetry to create massively disproportionate value.
- [Marginal-Returns Time Allocation Across Multiple Businesses](https://midas-wiki.vercel.app/wiki/frameworks/marginal-returns-time-allocation.md): Run several businesses at once by treating your day like a portfolio: work each one only up to the point where an extra hour stops paying, then move to the next.
- [Monoculture Diversification Framework](https://midas-wiki.vercel.app/wiki/frameworks/monoculture-diversification.md): Like monoculture farming leading to catastrophic crop failure (the Interstellar blight), relying 100% on one AI platform means 100% productivity loss during outages.
- [Multi-Model Workflow Patterns](https://midas-wiki.vercel.app/wiki/frameworks/multi-model-workflow.md): Using Gemini and Claude together — each model's strengths for different phases of the same project.
- [The Core Agent Loop (Observe-Think-Act)](https://midas-wiki.vercel.app/wiki/frameworks/observe-think-act-loop.md): Every AI agent runs a 3-step loop repeatedly until a "definition of done" is reached.
- [Opportunity-Cost Value Framing (Savings vs Revenue)](https://midas-wiki.vercel.app/wiki/frameworks/opportunity-cost-value-framing.md): Sell an automation on the revenue its freed time unlocks, not on the labor cost it saves — because the reallocated-revenue number is roughly 10x larger.
- [Platform Comparison: Claude vs Codex vs Gemini](https://midas-wiki.vercel.app/wiki/frameworks/platform-comparison.md): Strengths and weaknesses of each major AI coding platform. Key caveat: differences are only a few percentage points — all models are trained on the entire internet.
- [Posted-to-Production Time Ratio](https://midas-wiki.vercel.app/wiki/frameworks/production-to-post-time-ratio.md): The minutes of published content you ship divided by the minutes you spent producing it — the single number that decides whether you can reach content critical mass or not. [Source: Maker School — Resource Library Community Exclusives (p...
- [Prompt Contracts + Reverse Prompting](https://midas-wiki.vercel.app/wiki/frameworks/prompt-contracts.md): Vague tasks with no definition of done = the 1 problem with agent disillusionment. A prompt contract forces structured pre-execution agreement.
- [RACE Framework — Where to Apply AI in Any Business](https://midas-wiki.vercel.app/wiki/frameworks/race-framework.md): Nick's 4-stage filter for deciding what to automate first in a client business AND for filtering AI tool noise. Always start with the front-end revenue engine, never back-end ops.
- [Monthly Business Retrospective Ritual](https://midas-wiki.vercel.app/wiki/frameworks/retrospective-ritual.md): A recurring, written self-audit that turns raw activity into data-backed decisions — so you change strategy on evidence, not on a bad day's gut feeling. Run at two cadences: a weekly retro on individual lead-gen channels, and a monthly r...
- [Revenue-First Day Structure (Parkinson's Law)](https://midas-wiki.vercel.app/wiki/frameworks/revenue-first-day-structure.md): Order your workday by revenue proximity — outreach first, fulfillment last — because sales is the bottleneck and delivery time is elastic. Nick has run this same structure for ~1,200 days. [Source: Weekly Community Call 55]
- [Saturation Is Opportunity (Competitor-Quality Arbitrage)](https://midas-wiki.vercel.app/wiki/frameworks/saturation-is-opportunity.md): A "saturated" market is a buy signal, not a warning: competition proves there is money to be made, and in highly competitive knowledge-work markets the average competitor quality is very poor — so a little first-principles thinking makes...
- [Three Tiers of Skill Complexity](https://midas-wiki.vercel.app/wiki/frameworks/skill-complexity-tiers.md): A framework for deciding how to structure a skill based on what it needs to do.
- [Skill Debt Paydown](https://midas-wiki.vercel.app/wiki/frameworks/skill-debt-paydown.md): Early in a business you carry debt; one form is skill/technical debt — the things you don't yet know how to do. Optimize your earliest projects to pay it down as fast as possible, not to make money.
- [SOP-to-Skill Conversion Framework](https://midas-wiki.vercel.app/wiki/frameworks/sop-to-skill-conversion.md): Any existing SOP from your business can be converted into a skill. Even poorly written SOPs work as a starting point.
### Fulfillment
- [Two Recommended Agent Architectures for Client Work](https://midas-wiki.vercel.app/wiki/fulfillment/agent-architectures.md): Two proven architectures for structuring agent-based service delivery.
- [Client Servicing Flow](https://midas-wiki.vercel.app/wiki/fulfillment/client-servicing-flow.md): The complete end-to-end process from winning a project to retention. Tested from $0 to $72K/month. Nick's process — "inoculates against 95% of all agency problems."
- [Human-in-the-Loop Decision Framework](https://midas-wiki.vercel.app/wiki/fulfillment/human-in-the-loop.md): When to step in and when to let the agent run autonomously.
- [Kickoff Call SOP](https://midas-wiki.vercel.app/wiki/fulfillment/kickoff-call-sop.md): Structured onboarding call that eliminates 90% of future client problems. 15-20 minutes. Do everything live on the call so you walk away with everything you need.
- [Operational Playbook (Fix 90% of Agency Problems in 30 Days)](https://midas-wiki.vercel.app/wiki/fulfillment/operational-playbook.md): Five core operational problems and their fixes. If your agency is struggling, it's almost certainly one of these.
- [Project Delivery](https://midas-wiki.vercel.app/wiki/fulfillment/project-delivery.md): How to deliver a completed project — the three deliverables, the payment conversation, and the retainer upsell.
- [Project Management Systems: Status-Based vs Sub-Task-Based](https://midas-wiki.vercel.app/wiki/fulfillment/project-management-system.md): Two ways to structure agency delivery inside a PM tool (ClickUp, Monday, Trello, Basecamp, Asana): status-based, where a record moves through linear stages each owned by a person or department, and sub-task-based, where one parent task h...
- [Subcontracting Fulfillment: Payment Models + Full-Context Handoff](https://midas-wiki.vercel.app/wiki/fulfillment/subcontracting-delivery-economics.md): When you hand the back-end build of a signed project to a contractor (e.g. a Maker School member acting as back-office staff), two things determine whether it works: how you pay them, and how much context you give them at handoff. Nick's...
- [Team Resource Hub / Onboarding Library](https://midas-wiki.vercel.app/wiki/fulfillment/team-resource-hub.md): A single, role-gated internal knowledge base that holds every SOP, policy, and onboarding doc your team needs — so hiring, onboarding, and day-to-day questions run without you. Nick Sarayev calls it the thing that broke his scaling ceili...
- [Workflow Iteration Process](https://midas-wiki.vercel.app/wiki/fulfillment/workflow-iteration.md): How to take a workflow from first build to battle-tested reliability.
- [Client Workspace Organization](https://midas-wiki.vercel.app/wiki/fulfillment/workspace-organization.md): How to structure your file system for multi-client agency work.
### Leadgen
- [AI-Avatar Content Production Stack + When It Works](https://midas-wiki.vercel.app/wiki/leadgen/ai-avatar-content-production.md): Cloning a person's face and voice to generate video content without filming. Nick tested a specific stack and found it works only in narrow conditions: it fails wherever the audience knows you and the content is long, and it works only w...
- [AI's Role in Cold Email Copywriting](https://midas-wiki.vercel.app/wiki/leadgen/ai-role-in-copywriting.md): Nick's strong anti-AI stance for core copy, backed by 10,000+ data points from Maker School. Three approved narrow uses.
- [Campaign Iteration Strategy](https://midas-wiki.vercel.app/wiki/leadgen/campaign-iteration.md): Outbound is a data scientist game, not a creative writing game. You almost never one-shot a great campaign.
- [Cold-Call Number Deliverability & Caller Reputation](https://midas-wiki.vercel.app/wiki/leadgen/cold-call-number-deliverability.md): Cold-calling numbers carry a carrier-assigned reputation the same way sending domains carry an email reputation. A number can ring through, get flagged "scam likely," or be blocked outright depending on its score — and, like an email dom...
- [Cold Email Auto Research](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-auto-research.md): Applying the frameworks/auto-research loop to cold email optimization.
- [Cold Email Copywriting — The $15M Framework](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-copywriting.md): Nick's four-step framework that generated $15M+ in outbound sales, grounded in behavioral neuroscience and Robert Cialdini's "Influence."
- [Cold Email Deliverability (2026)](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-deliverability.md): Everything about getting cold emails into inboxes. The foundation: "whatever spammers do, I do the opposite."
- [Cold Email On Behalf of Clients — Compliance & Setup](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-for-clients.md): Running cold email for a paying client is mechanically identical to running it for yourself, but the risk profile changes: it's "no longer just on your ass." This page covers Nick's compliance framing (GDPR / CAN-SPAM gray area), how to...
- [Cold Email Master Playbook 2026 (Nick Saraev — Maker School Wrapped)](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-master-playbook-2026.md): ---
- [Cold Email Offers](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-offers.md): The difference between a campaign WITH a good offer vs WITHOUT: 5-10x reply rate. Not exaggerating — tested across hundreds of campaigns. Maker School data confirms the upper bound: as high as 10x [Source: Maker School — Building an Offer].
- [Cold Email Plausible Deniability — Hide the Source](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-plausible-deniability.md): "Found you on LinkedIn" is dead. Strip all source attribution from openers. Prospect should not know where you found them. They could be a fan, a vendor, a podcast host, a past contact.
- [Cold Email Scaling Infrastructure — 300 → 3,000+/day](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-scaling-infrastructure.md): Step-by-step process for scaling cold email from 300 to 3,000+ emails/day without burning deliverability or wasting mailbox capacity.
- [Cold Email Templates & Sequences](https://midas-wiki.vercel.app/wiki/leadgen/cold-email-templates.md): 5 proven templates, sequence structure, scheduling, and the pre-send checklist. Every template follows the four-part formula (Personalization → Who Am I → Offer → CTA).
- [Cold Outreach Live Build (Real Campaign, Real Results)](https://midas-wiki.vercel.app/wiki/leadgen/cold-outreach-live-build.md): A complete cold email campaign built from scratch with real metrics. Nick builds the system live, launches it, and shares results 13 hours later.
- [Community-Warming Lead Generation (Skool)](https://midas-wiki.vercel.app/wiki/leadgen/community-lead-generation.md): A lead-gen channel where you join niche communities (Skool, Facebook groups, Slack/Discord, Telegram, WhatsApp), post simple automation solutions to problems the community faces, and "warm" each community through daily manual engagement...
- [Grey Hat Outreach Techniques](https://midas-wiki.vercel.app/wiki/leadgen/grey-hat-outreach.md): Techniques Nick has used or observed. He does not explicitly recommend them — accounts have been suspended. Use at own risk.
- [Website Scraping via HTTP](https://midas-wiki.vercel.app/wiki/leadgen/http-scraping.md): The cheapest and fastest method for lead data extraction at scale.
- [Lead Scraping Tools & Pipelines (2025)](https://midas-wiki.vercel.app/wiki/leadgen/lead-scraping-tools-2025.md): Current tool stack and three proven scraping pipelines. Apollo + Apify scraper is no longer available. LeadsRapidly, Ample Leads, Export Apollo all "take forever" now.
- [LinkedIn Outreach Flow — 5-Step Sequence](https://midas-wiki.vercel.app/wiki/leadgen/linkedin-outreach-flow.md): Complete LinkedIn outbound sequence used by Nick. Built on top of a personal brand base, not as a standalone cold channel.
- [Multi-Platform Outreach Optimization](https://midas-wiki.vercel.app/wiki/leadgen/multi-platform-outreach.md): Every outreach platform has specific optimization levers. Master these to stand out in each channel's inbox.
- [Building a Personalization System](https://midas-wiki.vercel.app/wiki/leadgen/personalization-system.md): How to build AI personalization that actually works. The litmus test: "would a real person send this?" If your icebreaker sounds like something you'd text a friend, you're in good shape.
- [Podcast-Host Prospecting via LinkedIn Keyword Filters](https://midas-wiki.vercel.app/wiki/leadgen/podcast-host-prospecting.md): How to build an outreach list of podcast hosts who already own the audience of a target niche — Nick's worked example is service-based blue-collar business owners. The people who host niche podcasts have direct access to the exact buyers...
- [Post-Reply Booking Flow — Minimize Steps to Booked Call](https://midas-wiki.vercel.app/wiki/leadgen/post-reply-booking-flow.md): What happens after a positive cold email reply matters more than you think. Each extra message leaks ~50% of remaining prospects.
- [Search-Intent Scraping](https://midas-wiki.vercel.app/wiki/leadgen/search-intent-scraping.md): Search-intent scraping means sourcing prospects who already have a defined need, then reaching only that pre-qualified subset — instead of blasting a whole population cold. The signal that someone has the need does the qualifying for you...
- [Small TAM Tactics — Multi-Pass List Recycling](https://midas-wiki.vercel.app/wiki/leadgen/small-tam-tactics.md): Most operators assume a TAM is a one-shot resource. It's not. Same list, different mailboxes, different angles, no company name = 3× usable test points from one list.
- [Cold Email Subject Lines](https://midas-wiki.vercel.app/wiki/leadgen/subject-lines.md): The subject line's job is NOT to tell them everything. It's to make them click.
- [Upwork Evolution — How Strategy Changed 2021 → 2026](https://midas-wiki.vercel.app/wiki/leadgen/upwork-evolution-2025.md): Upwork landscape shifted significantly. Old tactics (high-volume templated apps) no longer work post-boosting. New optimal: one high-quality app + boost.
- [Video Outreach Production Guide](https://midas-wiki.vercel.app/wiki/leadgen/video-outreach-production.md): How to shoot, frame, and package a personalized outreach video (Loom-style) so it closes more than a raw screen recording. Covers gear, framing, on-screen proof tabs, annotation, the transcript-to-asset upsell, and an expected-value case...
### Sales
- [Age and Credibility in Selling](https://midas-wiki.vercel.app/wiki/sales/age-and-credibility-in-selling.md): Perceived age is a credibility signal buyers apply before they weigh your actual argument. Young or baby-faced sellers face real, unfixable skepticism, so the move is not to fake gravitas but to choose channels that route around the visu...
- [Calendar Optimization — Don't Open All Day](https://midas-wiki.vercel.app/wiki/sales/calendar-optimization.md): Block your calendar for deep work. Lose 1-2 meetings, gain 40% effectiveness. Net positive.
- [Case Study Tier List — Use What You Have](https://midas-wiki.vercel.app/wiki/sales/case-study-tier-list.md): You don't need the perfect case study to win. Each lower tier still provides social proof. Stop waiting for Tier 1 — use what you have.
- [Charm Pricing & The Round-Number Negotiability Tell](https://midas-wiki.vercel.app/wiki/sales/charm-pricing-round-number-tell.md): A round-number price like $1,000 signals to the buyer that you "just picked a thousand cuz it sounded nice" — that the price is arbitrary. An oddly specific number like $925 reads as deliberately calculated and defensible. Nick Saraev's...
- [Client Psychology](https://midas-wiki.vercel.app/wiki/sales/client-psychology.md): What clients actually care about and how to sell to that reality. The gap between what agencies think matters (tech) and what clients buy (outcomes) is where most deals die.
- [Cold Calling & Door-to-Door — Mindset and Geographic Arbitrage](https://midas-wiki.vercel.app/wiki/sales/cold-calling-door-to-door.md): Nick's early client acquisition ran on in-person cold calling and door-to-door pitching before any cold email or Upwork existed. The lasting value was two things: a rejection-hardened work ethic that makes every later outreach channel fe...
- [Cold Email Pricing Models — 5 Structures](https://midas-wiki.vercel.app/wiki/sales/cold-email-pricing-models.md): 5 pricing models for cold email / outbound services with selection logic. Risk tolerance + client fit determines which to use.
- [Renegotiating Your Comp/Equity From a Growth Frame](https://midas-wiki.vercel.app/wiki/sales/comp-renegotiation-from-growth.md): How to ask for more money, upside, or equity from a client or employer without triggering the reflex that kills the ask. The core move: never frame the request as recovering a loss ("you're underpaying me, pay up or I walk") — frame it a...
- [Discovery Call Script](https://midas-wiki.vercel.app/wiki/sales/discovery-call-script.md): The 30-45 minute sales call structure. People taught with this framework have closed deals up to $60K upfront.
- [Done-Work-First Pitch (The Autonomy Tier List)](https://midas-wiki.vercel.app/wiki/sales/done-work-first-pitch.md): A mental model for pitching anyone who makes more money than you: rank your offer by how much work it removes from the buyer's plate, not by how cheap or eager you are. The less they have to do, the higher the tier.
- [Hourly-to-Retainer Conversion Pitch](https://midas-wiki.vercel.app/wiki/sales/hourly-to-retainer-conversion.md): The exact pitch for moving an existing hourly client onto a flat monthly retainer. Nick's framing: a retainer is not a price increase, it is an incentive mechanism that takes every negative of hourly work and flips it to positive for bot...
- [Industry Targets for AI Automation (2025)](https://midas-wiki.vercel.app/wiki/sales/industry-targets.md): 5 industries that will pay premium for AI automation, with specific systems to sell each. The hidden lever: margin math.
- [MRR Definition — What Counts and What Doesn't](https://midas-wiki.vercel.app/wiki/sales/mrr-definition.md): True MRR = revenue with CONTRACTUAL future promise of recurrence. Common mistake: inflating numbers with "expected" repeat business.
- [Newswire / Paid PR for "As Seen In" Social Proof](https://midas-wiki.vercel.app/wiki/sales/newswire-pr-social-proof.md): Buying a paid press-release blast ("newswire") to earn "as seen in" / "featured in" logos from major outlets, then displaying those logos on your agency site to raise perceived authority so you can charge more and close more. This is the...
- [Niche Selection for Cold Outreach](https://midas-wiki.vercel.app/wiki/sales/niche-selection.md): Three criteria Nick always uses, plus a rapid problem discovery method. Total time: 3 minutes. "Don't spend more than 20 minutes. The market is a better tutor than analysis."
- [Objection Handling & Closing Mechanics](https://midas-wiki.vercel.app/wiki/sales/objection-handling-and-closing.md): The tactical layer of a sales call after value has been established: how to deliver price, isolate the real objection, sequence the ask, and close on the call. Drawn from Maker School sales training (Doug Shankman on price delivery and t...
- [Offer Construction Formula](https://midas-wiki.vercel.app/wiki/sales/offer-construction.md): Every cold outreach offer contains three components: a problem to solve, a solution shape (you don't need a built product), and a time-bound guarantee.
- [Perceived Value vs Real Value in Commoditized Markets](https://midas-wiki.vercel.app/wiki/sales/perceived-value-positioning.md): When products in a market are objectively similar, you don't win on the product — you win on how clearly you communicate its value. Perceived value is "which is all marketing," and it's entirely up to how you shape the buyer's perception...
- [Pre-Call Show Rate Tactics — Two-Message Sequence](https://midas-wiki.vercel.app/wiki/sales/pre-call-show-rate-tactics.md): Two-reminder sequence + humanization hack. Anecdotally raises show rate across multiple agencies tested.
- [8 Agency Pricing Methods](https://midas-wiki.vercel.app/wiki/sales/pricing-methods.md): Exhaustive guide to pricing automation services. Tested from $0 to $72K→$100K+/month. No pricing method is perfect — each has trade-offs. The best agencies constantly re-evaluate.
- [Pricing Speed Framing — Fix the "Fast = Cheap" Problem](https://midas-wiki.vercel.app/wiki/sales/pricing-speed-framing.md): "I can do it in 10 minutes" triggers "then I'll pay $100" thinking from small clients. Two-track solution.
- [Project Scoping Method — 15-Min Feasibility Check](https://midas-wiki.vercel.app/wiki/sales/project-scoping-method.md): How to scope an automation project before quoting. Data object mapping verifies feasibility without over-engineering.
- [Proof-of-Concept / Pilot Closing](https://midas-wiki.vercel.app/wiki/sales/proof-of-concept-pilots.md): Offering a paid or free trial pilot before the full engagement de-risks the deal for the buyer. Because they carry less risk, you can charge more — and you close on evidence (real leads, screenshots, reply threads) instead of promises. [...
- [Proposal Writing — Saraev Framework](https://midas-wiki.vercel.app/wiki/sales/proposal-writing.md): How to write proposals that close $10K+ clients. Source: Nick Sarayev / Maker School "How to write proposals that close $10K+ clients" (May 2026).
- [Reactivation Campaign](https://midas-wiki.vercel.app/wiki/sales/reactivation-campaign.md): A reactivation campaign is a manual re-pitch of past clients and positive leads that never closed. The goal is to recoup the sunk cost you already paid to acquire those contacts by extracting more work from them. You already spent money...
- [Retainer Deliverables — Line Items That Justify Premium](https://midas-wiki.vercel.app/wiki/sales/retainer-deliverables.md): What to include in an automation retainer beyond raw hours. Each line item raises perceived value AND reduces churn.
- [Retainer Scope Management — Handle Out-of-Scope Without Damage](https://midas-wiki.vercel.app/wiki/sales/retainer-scope-management.md): Two extremes fail: rigid scope policing (clients hate, leave) and saying yes to everything (you get exploited). Optimal: free goodwill on small asks + retainer pitch on repeat requests.
- [RevOps Offer Construction — Force Transactional Framing](https://midas-wiki.vercel.app/wiki/sales/rev-op-offer-construction.md): Pure consulting / RevOps framing underperforms cold lead-gen offers. Solution: force-fit RevOps into a transactional format by guaranteeing revenue impact in dollars and days.
- [Multi-Touch Sales Follow-Up Cadence](https://midas-wiki.vercel.app/wiki/sales/sales-follow-up-cadence.md): A stacked, multi-channel follow-up system for converting verbal interest into closed deals. The core insight: most prospects who said "not now" will roll into a buying window within a week or two, so you follow up persistently, across ch...
- [3 Sales Systems](https://midas-wiki.vercel.app/wiki/sales/sales-systems.md): Three distinct sales pipelines for different deal sizes. Choose based on average deal value and client sophistication.
- [Selling Cold Email Systems (Consultative Education)](https://midas-wiki.vercel.app/wiki/sales/selling-cold-email-systems.md): Connor Kaplan's method for closing cold email service deals: build trust by educating the prospect through your lens. You walk them through the full process flow — infrastructure → strategy → list building → copy → inbox management — and...
- [Service Positioning in the AI Age](https://midas-wiki.vercel.app/wiki/sales/service-positioning.md): How to position AI agency services for maximum defensibility and revenue.
- [AI Service Tier List (2025)](https://midas-wiki.vercel.app/wiki/sales/service-tier-list.md): 14 AI services ranked by revenue impact and sellability. S-tier = sell first, E-tier = avoid. Based on Nick's experience across hundreds of projects.
- [Upsell & LTV System — Engineering $30-40K Per Client](https://midas-wiki.vercel.app/wiki/sales/upsell-and-ltv-system.md): 90% of agency money comes from retained relationships, not acquisition. Nick's average LTV target: $30-40K per client via planned upsell mechanics. NOT random upsells — engineered from the first call.
### Strategy
- [Software Affiliate Links as an Operating-Margin Lever](https://midas-wiki.vercel.app/wiki/strategy/affiliate-revenue-margin-lever.md): Every AI automation agency signs its clients up to a stack of third-party software during delivery. Routing those signups through your own affiliate links turns a cost center into a passive, lifetime revenue stream — and Nick frames it a...
- [Agency Day-One Bootstrap: Name, Domain, Email](https://midas-wiki.vercel.app/wiki/strategy/agency-bootstrap-setup.md): The Day-1 setup that turns "I want to start an agency" into a working business identity: an operating name, a domain, and a primary business email. Nick frames the whole block under a meta-rule — aim for 80% not 100% ("this is the gym, n...
- [Agency Business Model](https://midas-wiki.vercel.app/wiki/strategy/agency-business-model.md): How Nick structures and runs LeftClick, his AI automation agency.
- [AI Operating System — Premium Upsell Product](https://midas-wiki.vercel.app/wiki/strategy/ai-os-product.md): "AI OS" = a package of 5-6 skills wrapped in a dashboard + chat window. Premium upsell for existing clients. NOT a cold-outbound offer — too broad to convert.
- [AI — What NOT to Automate](https://midas-wiki.vercel.app/wiki/strategy/ai-what-not-to-automate.md): Wiki covers what to automate extensively. This is the inverse: what NEVER to automate, with a clear decision rule.
- [Why NOT to Sell to Local Businesses](https://midas-wiki.vercel.app/wiki/strategy/avoid-local-businesses.md): Counterintuitive advice from someone making $170K/month: local businesses are a trap for AI automation agencies. "The further away I've gotten from local businesses, the more money I've made."
- [Building on Hidden / Unofficial APIs](https://midas-wiki.vercel.app/wiki/strategy/building-on-hidden-apis.md): Building a SaaS or automation product on top of someone else's public, hidden, or unofficial API — or on cookie-based browser automation of their platform — is a fragile Jenga tower. The platform periodically changes its API specifically...
- [Claude Cowork Enterprise Rollout + AI-Training Retainer](https://midas-wiki.vercel.app/wiki/strategy/claude-cowork-enterprise-training.md): An agency offer for selling AI enablement to non-technical mid-market and enterprise teams: roll them onto Claude Cowork (the non-developer version of Claude Code), train the staff on how agents actually work, build custom skills for the...
- [Client Concentration Risk and the Small-Account Hedge](https://midas-wiki.vercel.app/wiki/strategy/client-concentration-risk.md): Concentration risk is the exposure that comes from earning most of your revenue from a tiny number of clients. When roughly 2 clients account for the bulk of revenue, a single departure can wipe out around 40% of the business overnight....
- [Community / Info-Product Annual Discount Promo](https://midas-wiki.vercel.app/wiki/strategy/community-annual-promo.md): A periodic, time-boxed annual-plan discount offered to a paid community (Skool) or info-product — the correct way to run a price promotion without churning your existing members. Instead of cutting the base monthly price (which punishes...
- [Content Signal vs Noise — Block 99%, Mine the 1%](https://midas-wiki.vercel.app/wiki/strategy/content-signal-vs-noise.md): 99% of AI news, social feeds, and trend content is noise. Miss the 1% signal occasionally; skip 99% noise constantly. Net positive.
- [Content Strategy for Agency Builders](https://midas-wiki.vercel.app/wiki/strategy/content-strategy-for-agencies.md): Content is an inbound channel that activates AFTER outbound is maxed out, not before. Wrong order = wasted years + low-quality leads.
- [Distribution Methods & the 5/95 Rule](https://midas-wiki.vercel.app/wiki/strategy/distribution-methods.md): Software can no longer be a moat. The moat is in distribution — your ability to reach and retain people.
- [Enterprise vs SMB Positioning — Binary Decision Rule](https://midas-wiki.vercel.app/wiki/strategy/enterprise-vs-smb-positioning.md): Small company = add revenue. Large company = maximize margins. Switch the value proposition completely or you'll fail.
- [Foreign-Market Language as an AGI-Proof Trust Signal](https://midas-wiki.vercel.app/wiki/strategy/foreign-market-language-trust.md): Learning the buyer's language is the highest-ROI trust earner for a non-native seller entering a foreign market. Nick's framing: some signals earn immediate respect because they visibly cost time and effort to acquire — being "naturally...
- [Founder Focus System](https://midas-wiki.vercel.app/wiki/strategy/founder-focus-system.md): Focus is the raw input a founder converts into money: either your focus converts to your dollars, or others — short-form content creators — take your focus and convert it to theirs. Nick frames focus as a trainable capacity across three...
- [Focus on Front-End Business Activities](https://midas-wiki.vercel.app/wiki/strategy/front-end-focus.md): Nick's core thesis: The difference between people who make money with AI and those who collect shiny objects is where they focus. Winners focus on the front end of the business — the revenue-generating activities.
- [Geographic Market Selection (Friction vs Earning Power)](https://midas-wiki.vercel.app/wiki/strategy/geographic-market-selection.md): Which country's clients you sell to is a tradeoff between earning power (how much they pay) and friction (how hard the close is). Nick's answer: don't automatically chase the highest-paying markets — run the friction-reduction ratio agai...
- [Hire Last — Don't Hire Until $20-25K/mo](https://midas-wiki.vercel.app/wiki/strategy/hire-last-principle.md): Hiring at $5K/mo scales inefficiency, not business. Fix the front-end bottleneck (lead gen, sales) first. AI grows exponentially; humans grow ~1%/year.
- [Horizontal Leverage](https://midas-wiki.vercel.app/wiki/strategy/horizontal-leverage.md): Don't think "automate 100% of 1 role" (= 1 unit of value). Think "automate 90% of 10,000 roles" (= 9,000 units of value).
- [Hourly Rate Decision Matrix — Scale to $100/hr Without Hiring](https://midas-wiki.vercel.app/wiki/strategy/hourly-rate-decision-matrix.md): Practical scaling rule for time-capped solopreneurs at $10-20K/mo. Use true hourly rate as a decision gate on every new opportunity.
- [Input Goals vs Output Goals — Personal Goal-Setting System](https://midas-wiki.vercel.app/wiki/strategy/input-goals-vs-output-goals.md): Replace output goals (revenue, subscribers) with input goals (daily actions). Output goals plateau-near; input goals overshoot.
- [Lead Gen Channel Progression — Saturation Story](https://midas-wiki.vercel.app/wiki/strategy/lead-gen-channel-progression.md): Each channel saturates at a revenue ceiling. "What got you here won't get you there." Nick's documented path through channels with real revenue bands.
- [Local Models vs Frontier — the Token-Subsidy Arbitrage](https://midas-wiki.vercel.app/wiki/strategy/local-models-vs-cloud-arbitrage.md): Running open-source/local models on your own hardware feels like a cost win, but it forfeits three things at once: frontier-level intelligence, cloud-scale parallelism, and a temporary pricing subsidy that makes very smart models absurdl...
- [Momentum & Consistency Before Your First Client](https://midas-wiki.vercel.app/wiki/strategy/momentum-before-first-client.md): Before you land client 1, the single most crucial part of acquiring your first customer is momentum, and you only build momentum by showing up daily for a long period. Consistency is a trainable, cornerstone skill — and before you have a...
- [Money Model — Three-Tier LTV Funnel](https://midas-wiki.vercel.app/wiki/strategy/money-model-three-tier.md): Three money models for the same client acquisition. Higher tier = higher LTV without proportional CAC increase. Same lead becomes 10× more valuable.
- [n8n vs Claude Code — Decision Guide](https://midas-wiki.vercel.app/wiki/strategy/n8n-vs-claude-code-decision.md): Practical guidance on when to build in n8n, when in pure Claude Code, and when to hybrid. Default: use what you're best at. Don't kneecap yourself by chasing the popular tool.
- [Niche Pack Strategy](https://midas-wiki.vercel.app/wiki/strategy/niche-pack-strategy.md): A niche pack is a curated bundle of 2-4 interlocking AI-automation systems that already sell well to one specific niche, shipped as 80%-done templates — systems Nick and Maker School community members have personally sold. The strategy:...
- [No-Code vs Agent Stack — 3-Tier Delivery Architecture](https://midas-wiki.vercel.app/wiki/strategy/no-code-vs-agent-stack.md): Three-tier model for how to deliver automation in 2026. Most builders fit into one tier; the sweet spot is Tier 2.
- [Skip Personal Branding Early — Do Invisible Direct Outreach](https://midas-wiki.vercel.app/wiki/strategy/outreach-over-branding-early.md): When starting an agency, Nick's default recommendation is to avoid personal branding entirely and do direct outreach instead. Brands take a long time to pay off, and a public brand can get you judged by peers — a real cost if you're job-...
- [Partnership / Piggyback Distribution Model](https://midas-wiki.vercel.app/wiki/strategy/partnership-distribution.md): Instead of building your own lead-gen distribution from scratch, you piggyback on people who ALREADY have an audience: cold-email operators with pre-existing distribution and offer them a revenue share to fulfill AI/automation projects f...
- [Payment-Processor Resilience](https://midas-wiki.vercel.app/wiki/strategy/payment-processor-resilience.md): Treating Stripe (or any single payment processor) as a single point of failure is one of the most under-priced risks an agency carries. Processors freeze or pull accounts with your money still inside — including on payroll day — and the...
- [Productization Method: Reverse-Engineering Custom Builds Into Products](https://midas-wiki.vercel.app/wiki/strategy/productization-method.md): How to turn a repeatable custom build into a productized system by working backward from what the client actually sees — deliverables → tasks → job roles → cost → price. Nick frames this reverse-engineering / back-propagation as "at the...
- [Profit First — Designing Margins Instead of Inheriting Them](https://midas-wiki.vercel.app/wiki/strategy/profit-first-margin-design.md): Decide your target profit percentage first, then force every expense to fit inside the complement — instead of letting profit be whatever's left after costs "happen to you."
- [Runway & Self-Funding — Don't Bootstrap From Zero](https://midas-wiki.vercel.app/wiki/strategy/runway-and-self-funding.md): Do not start an agency hoping the business rescues you from an empty bank account. Nick's stated 1 regret is starting entrepreneurship broke: get a cheap job first, build runway, then funnel that income into outreach to spin up the flywh...
- [SaaS is Dying — Distribution is the New Moat](https://midas-wiki.vercel.app/wiki/strategy/saas-is-dying.md): When anyone can build Netflix in 5 minutes with 3-4 agents, software quality is no longer a moat.
- [Scaling Framework (4 Steps with Revenue Ceilings)](https://midas-wiki.vercel.app/wiki/strategy/scaling-framework.md): The progression from custom projects to scalable business. Each step has a natural revenue ceiling — you must evolve to break through. Based on Nick's path from $500/project to $170K/month.
- [Value-Based Pricing (VBP)](https://midas-wiki.vercel.app/wiki/strategy/value-based-pricing.md): Price based on value delivered, NOT costs or competitor pricing.
### Technical
- [AI Agent Evaluation & Monitoring at Scale](https://midas-wiki.vercel.app/wiki/technical/agent-evaluation-monitoring.md): How to track whether a deployed AI agent is actually working — at the scale of tens of thousands of conversations — and the sharper question of what "working" even means. Nick's answer separates two concerns: a technical self-verificatio...
- [Agent Harness Architecture](https://midas-wiki.vercel.app/wiki/technical/agent-harness.md): An agent harness = everything that wraps around the LLM that is NOT the LLM itself. The harness is the barrel of the gun; the LLM is the gunpowder.
- [Claude Routines — Anthropic's n8n Replacement](https://midas-wiki.vercel.app/wiki/technical/claude-routines.md): Anthropic's Claude Routines = scheduled / triggered agentic workflows. Replaces n8n/Make for many use cases. Higher cost than n8n but zero drag-and-drop and leverages Anthropic infra directly.
- [Cloud Deployment Architecture](https://midas-wiki.vercel.app/wiki/technical/cloud-deployment.md): Key principle: upload only execution scripts (deterministic parts), NOT the orchestrator. The agent is the architect; the cloud workflow is the building.
- [Deployment Options](https://midas-wiki.vercel.app/wiki/technical/deployment-options.md): Three primary deployment targets for AI agency work.
- [EU / GDPR Compliance Handling for AI Agencies](https://midas-wiki.vercel.app/wiki/technical/eu-gdpr-compliance-handling.md): How Nick handles data-protection demands from EU clients when building AI systems that pass user data (names, emails, PII) through third-party model APIs. His core stance: do not lead with compliance, engineer around it technically, and...
- [Gemini Antigravity IDE](https://midas-wiki.vercel.app/wiki/technical/gemini-antigravity.md): Google's AI-native IDE for vibe coding. Key features, workflows, and how Nick uses it alongside Claude Code.
- [Internet Automation Spectrum (3 Levels)](https://midas-wiki.vercel.app/wiki/technical/internet-automation-spectrum.md): Three levels of internet automation, from cheapest/fastest to most expensive/universal.
- [MCP Servers (Model Context Protocol)](https://midas-wiki.vercel.app/wiki/technical/mcp-servers.md): MCPs = skills that other people/developer teams make for you. They extend Claude Code's capabilities with third-party integrations.
- [Mobile Stack: Expo + React Native (+ Supabase)](https://midas-wiki.vercel.app/wiki/technical/mobile-stack-expo-react-native.md): Preferred Claude-Code mobile app stack from Nick Sarayev's 2026 mobile course. Single codebase ships to iOS, Android, and web. Cloud builds via Expo EAS, hot reload on device, huge ecosystem.
- [Prompt Engineering (14 Techniques)](https://midas-wiki.vercel.app/wiki/technical/prompt-engineering.md): Production prompt engineering techniques from $2.4M+ in prompt-driven revenue. Not theory — tested across 1,000+ business prompts.
- [RAG Anti-Hallucination — Cohere Reranker Pattern](https://midas-wiki.vercel.app/wiki/technical/rag-anti-hallucination.md): Production RAG setup that minimizes hallucination. Key piece: Cohere's reranker between vector retrieval and LLM context.
- [Security Checklist for AI Agent Systems](https://midas-wiki.vercel.app/wiki/technical/security-checklist.md): "Everything on planet Earth is hackable. It's always just a question of how hackable." Cover the low-hanging fruit that eliminates 90% of attackers.
- [Sub-Agent Architecture](https://midas-wiki.vercel.app/wiki/technical/sub-agent-architecture.md): Skills can spawn sub-agents for parallel work, creating a parent-child orchestration pattern.
- [Vibe Coding Security](https://midas-wiki.vercel.app/wiki/technical/vibe-coding-security.md): The 5 specific vulnerabilities that cover 80% of vibe-coded app security issues, plus the Stripe integration pattern.
### Workflows
- [Agent Teams](https://midas-wiki.vercel.app/wiki/workflows/agent-teams.md): Claude Code's built-in feature for orchestrating multiple agents in a unified interface.
- [Cross-Agent Task Continuity](https://midas-wiki.vercel.app/wiki/workflows/cross-agent-task-continuity.md): A pattern for letting a second agent pick up a build exactly where the first left off when the first runs out of usage or context mid-task. The mechanism is a single shared file every agent reads and writes, so the handoff carries no dep...
- [Debate Pattern (Multi-Round Agent Discussion)](https://midas-wiki.vercel.app/wiki/workflows/debate-pattern.md): Agents see and respond to each other's outputs across multiple rounds, producing increasingly nuanced solutions.
- [Fan Out / Fan In Research Pattern](https://midas-wiki.vercel.app/wiki/workflows/fan-out-fan-in.md): Spawn N cheap research sub-agents in parallel, then synthesize with one powerful agent.
- [Implement-Review-Resolve Pattern](https://midas-wiki.vercel.app/wiki/workflows/implement-review-resolve.md): Three separate agents with ZERO shared context for maximum quality.
- [MCP-to-Skill Conversion Pattern](https://midas-wiki.vercel.app/wiki/workflows/mcp-to-skill-conversion.md): Use MCPs for quick proof-of-concept, then convert to skills for production. Saves 50-100x in context tokens.
- [Meta-Directives (Chaining Workflows)](https://midas-wiki.vercel.app/wiki/workflows/meta-directives.md): After building individual workflows, chain them with a parent directive that orchestrates the full pipeline.
- [Mobile Testing Loop (3-Tier)](https://midas-wiki.vercel.app/wiki/workflows/mobile-testing-loop.md): Mandatory 3-tier testing protocol for any mobile app built with Claude Code. Each tier catches bugs the others can't. Skipping tiers = bugs in production. Source: Nick Sarayev's 2026 mobile course.
- [Pipeline Pattern (Sequential Specialist Handoff)](https://midas-wiki.vercel.app/wiki/workflows/pipeline-pattern.md): Chain specialized agents sequentially, each handling one phase of the workflow.
- [Runnable Agent Loops (copy-paste /loop prompts)](https://midas-wiki.vercel.app/wiki/workflows/runnable-agent-loops.md): Three ready-to-run loop recipes for coding-agent harnesses (Claude Code, Codex, and similar) that turn a plain prompt into an autonomous background worker. The unlock is not a smarter model: it is the loop plus a few strategic constraint...
- [Social Media Automation Patterns](https://midas-wiki.vercel.app/wiki/workflows/social-media-automation.md): Tool-agnostic architecture patterns extracted from Nick's n8n blueprints, applicable to any automation platform (Claude Code skills, Modal, etc.).
- [Stochastic Multi-Agent Consensus](https://midas-wiki.vercel.app/wiki/workflows/stochastic-consensus.md): Because LLMs are stochastic (don't return same answer twice), running N agents on the same query yields more total unique answers than running one agent N times serially.
- [Task-Do-Verify Loop](https://midas-wiki.vercel.app/wiki/workflows/task-do-verify-loop.md): The 1 reason people get bad results with AI: they give Claude a task, it does the task, they give another task. No verification.
- [Vibe Coding Design Loop](https://midas-wiki.vercel.app/wiki/workflows/vibe-coding-loop.md): The 8-step process for building full-stack apps conversationally with AI. Order matters — especially auth at the end.

## AI in Ecommerce (10M+ PLN)

Applying AI inside larger ecommerce operations: the knowledge layer, team systems, topics and case sources for Polish ecom prospects and clients.

Full bundle: [ai-ecom.txt](https://midas-wiki.vercel.app/kb/_full/ai-ecom.txt)

- [AI Ecom Playbook — Index](https://midas-wiki.vercel.app/kb/ai-ecom/index.md): Content catalog. One line per page. Update on every ingest.
- [Log](https://midas-wiki.vercel.app/kb/ai-ecom/log.md): Chronological, append-only. Format: [YYYY-MM-DD] action | title then 1-line summary + pages touched.
- [AI in Ecommerce — Evolving Thesis (2026)](https://midas-wiki.vercel.app/kb/ai-ecom/overview.md): This page is the rolling synthesis. It changes as sources are ingested. Every revision should make the thesis sharper, not longer.
### People
- [Alex Hormozi](https://midas-wiki.vercel.app/kb/ai-ecom/people/alex-hormozi.md): Founder of Acquisition.com (operator-focused investment + advisory firm) and ACQ (the membership / advisory practice). Co-founder + content creator behind a large business-education audience. Reports his portfolio companies generating ~$...
- [Amit Pasha](https://midas-wiki.vercel.app/kb/ai-ecom/people/amit-pasha.md): Founder/CEO of Rich Panel — customer service software for ecommerce brands. 37 years old. Sequoia-backed (early stage; said no to subsequent rounds after becoming profitable). 18 months ago pivoted toward agentic-future positioning while...
- [Cody Plofker](https://midas-wiki.vercel.app/kb/ai-ecom/people/cody-plofker.md): CMO at Jones Road Beauty (premium beauty brand, owned-retail-bound, recently launched wholesale). Co-host of The Operators podcast and Marketing Operators podcast (with Connor MacDonald + others).
- [Connor MacDonald](https://midas-wiki.vercel.app/kb/ai-ecom/people/connor-macdonald.md): CMO at Hexclad (premium cookware). Previously associated with Ridge ecosystem. Co-host of Marketing Operators podcast (with Cody Plofker). Active Higgsfield body-swap experimenter; "AI apologist" who's "always willing to be swayed" towar...
- [Craig Foster](https://midas-wiki.vercel.app/kb/ai-ecom/people/craig-foster.md): Former AI lead at Crocs / Hadude. Now founder of Chatwalrus (chatwalrus.com) — AI training program for "normal people" to become operationally fluent. Hosts a private weekly off-the-record CEO conversation (invite-only via existing-membe...
- [Dar Denny](https://midas-wiki.vercel.app/kb/ai-ecom/people/dar-denny.md): 10-year media buyer / creative strategist / performance creative team leader. Three current roles: (1) partner / co-founder of a small boutique agency, (2) consultant on creative operations, (3) DTC content creator on YouTube. Soon-to-be...
- [Katie (Caden Lane)](https://midas-wiki.vercel.app/kb/ai-ecom/people/katie-caden-lane.md): Founder of Caden Lane — DTC baby/maternity brand. Co-host on The Operators podcast (recurring guest). Nine-figure brand, DTC-only until recently launched wholesale; 20% of revenue from mobile app; planning toward owned retail.
- [Krishna (Saras Analytics)](https://midas-wiki.vercel.app/kb/ai-ecom/people/krishna-saraswathi.md): CEO/co-founder of Saras Analytics — a 10-year-old data warehouse + analytics company for ecommerce brands. 200-person team based partly in India. Sequoia-backed at early stage; declined later rounds after profitability.
- [Matt Bertulli](https://midas-wiki.vercel.app/kb/ai-ecom/people/matt-bertulli.md): Co-host of The Operators podcast. Founder/operator at Pilo (consumer brand referenced in Postscript ad-reads). DTC operator perspective in Operators episodes. Active Claude Code user — frequently brings the "I built X in Claude Code last...
- [Nik Sharma](https://midas-wiki.vercel.app/kb/ai-ecom/people/nik-sharma.md): DTC operator + investor. Founder of Sharma Brands. 9 years building, scaling, and investing in consumer brands. Hosts the Limited Supply podcast. Hosts the Ecom AI Summit in New York (Webster Hall, 400+ brands). Newsletter at nik.co.
- [Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/people/ron-shah.md): D2C operator turned advisor. Built Obvi from $0 to $100M+ (supplements / wellness DTC). Now advises 25-50+ DTC brands as a fractional growth consultant. Hosts the Chew On This podcast. Public-facing AI evangelism through ronshah.co/blog...
- [Sean Frank](https://midas-wiki.vercel.app/kb/ai-ecom/people/sean-frank.md): CEO of Ridge (premium men's wallets + accessories, 21 years in business). Co-host of The Operators podcast. Operator at scale: 600,000 wallet customers acquired in 2025 (US-only); 10M+ cumulative customers.
- [Taylor Holiday](https://midas-wiki.vercel.app/kb/ai-ecom/people/taylor-holiday.md): Founder of Common Thread Collective (CTC) — DTC marketing/finance agency. Sold half of CTC mid-2025. Works across ~200 brands at the Shopify mid-market scale ($5M-$50M).
### Sources
- [AI Content Detectors — Top Tools for Businesses](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2024-07-31_ai-content-detectors.md): - AI detection works on two signals: burstiness (sentence length variation) and perplexity (word-combination predictability). Human writing is bursty and unpredictable; default AI output is the opposite.
- [How to Humanize AI Content — Tips and Techniques (2025)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2025-05-06_how-to-humanize-ai-content.md): - 88% of marketers use AI; 93% of those use it for content. That much output guarantees that "default AI voice" is now the noise floor. Humanization is the only way to stand out.
- [AI Trends in 2026 — What Businesses Need to Know](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2025-07-08_ai-trends-2026.md): - Eight macro trends for 2026 ecom: generative content, personalization, conversational agents, predictive analytics, digital twins, AI-cybersecurity, multimodal AI, ethics/regulation.
- [AI Statistics — Key Trends and Data for 2026](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2025-09-26_ai-statistics.md): - Market size: Global AI market projected at $757.6B in 2025 → $1.8T by 2030 (19.2% CAGR). Ecom-AI specifically: $17B by 2030. NLP: $112B by 2030.
- [AI Images for Amazon That Convert Better Than Real Photos](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-01-15_ai-images-amazon-convert-better.md): - Phone photos → Amazon-listing-ready images in ~90 minutes via Nano Banana Pro. The bottleneck wasn't "good photography" — it was a 3-week wait that cost 3 weeks of testing velocity.
- [Building a Client/Team Brain with NotebookLM](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-01-20_guide-client-team-brain-notebooklm.md): - Agency teams suffer from scattered client info: emails, call notes, Slack, docs. NotebookLM centralizes it as a queryable Client Brain — preps calls in <5 min, onboards new team members in hours not weeks.
- [Running GPT Inside Your Spreadsheets for Bulk Tasks](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-01-27_guide-using-ai-in-excel.md): - Talarian's "GPT for Work" add-ons embed GPT inside Google Sheets and Excel. Three working modes — Agent (interactive), Bulk (consistent at scale), Functions (=GPT(...) formulas in cells).
- [AI-Powered UX Audit vs. Competitors (Gemini 3)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-10_guide-ai-powered-ux-audit-competitors.md): - A UX audit that used to cost thousands and take weeks now runs in a single Gemini 3 session. Compares your site against 3 competitors and outputs: competitive matrix + SWOT + Top 3 friction killers + 90-day roadmap.
- [Creating Brand-Consistent Infographics with Gemini Flash Image](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-17_guide-brand-consistent-infographics-gemini-flash.md): - "Create an infographic" prompts produce cluttered AI slop. The 5-step framework converts blog posts/reports/text into polished brand assets in ~15 minutes.
- [Use Claude in PowerPoint (Official Add-in)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-24_guide-claude-in-powerpoint.md): - Claude runs as an official Microsoft add-in inside PowerPoint — template-aware, meaning it reads slide masters, layouts, fonts, colors. Generates and reshapes slides without leaving the app.
- [I Replaced a $12K/Month Role With a Prompt — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-24_ronshah_i-replaced-12k-role-with-prompt.md): - Reporting analyst spending 3 days/week compiling Shopify + Meta + Klaviyo + affiliate data was replaced by an AI workflow Ron built in one afternoon.
- [My Morning Now Starts With Claude, Not Slack — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-25_ronshah_morning-starts-with-claude-not-slack.md): - Ron's daily routine: 6:15am Claude with yesterday's Shopify export + Meta dashboard + Klaviyo flow performance → "what changed? what should I care about? what would you do differently?"
- [Your Agency Doesn't Want You to Learn AI — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-26_ronshah_your-agency-doesnt-want-you-to-learn-ai.md): - Of 25+ brands Ron advises, almost all use agencies. Almost none of those agencies are teaching their clients how to use AI — because if the brand can self-generate 20 ad concepts in 10 minutes, what is the agency for?
- [I Found $50K in Dormant Revenue — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-27_ronshah_i-found-50k-in-dormant-revenue.md): - Affiliate partner spreadsheet: 163 partners across 4 tabs in inconsistent formats. Manual analysis would take a team member 2 weeks.
- [20 Ad Angles in 10 Minutes — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-02-28_ronshah_20-ad-angles-in-10-minutes.md): - A brand stuck in a "creative rut" — same angles, same hooks, declining CTR every month. Agency kept pitching "fresh creative" that looked like last quarter's "fresh creative."
- [Stop Hiring for Tasks, Start Building Systems — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-01_ronshah_stop-hiring-for-tasks-start-systems.md): - Default hiring instinct in DTC: "Need reports? Hire an analyst. Need creative? Hire a designer." Ron's reframe: before you hire, ask "is this a task or a judgment call?"
- [The 30-Day Reporting Cycle Should Be Illegal — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-02_ronshah_30-day-reporting-cycle-illegal.md): - Brands waiting for end-of-month review are flying blind. 30 days of bad spend can be the difference between a profitable quarter and a cash crisis.
- [Building a Viral TikTok Script Machine with AI](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-03_guide-viral-tiktok-script-machine-ai.md): - A 4-step pipeline that converts competitor TikTok analysis into a compounding script-generation engine: Genspark (downloads videos at scale) → Gemini (transcribes + structurally analyzes each) → pattern library (spreadsheet/Notion DB)...
- [AI Won't Replace Marketers — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-03_ronshah_ai-wont-replace-marketers.md): - Two camps both wrong: "AI replaces everyone" vs "AI can't do what we do." Reality on the ground: operators who use AI just move faster. No robot overlords, no mass unemployment — just a widening speed gap.
- [My Kids Will Never Know Marketing Without AI — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-04_ronshah_my-kids-will-never-know-marketing-without-ai.md): - Generational reframe: by the time Ron's kids work, marketing without AI will feel as weird as marketing without the internet feels today.
- [The Amazon Listing Image SOP Nobody's Talking About](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-05_amazon-listing-image-sop.md): - Most Amazon listing images are designed on vibes — what looks cool, what the competitor did, what a designer threw together. The principle being argued: flip the process — start with customer data, then design.
- [The 3-Layer AI Stack Every DTC Brand Needs — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-05_ronshah_3-layer-ai-stack.md): - After building AI systems for 50+ brands, Ron landed on a 3-layer model. Most brands have zero. A few have one. Almost none have all three.
- [Fix Your AI Image Quality (Nano Banana 2 vs Pro)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-10_guide-fix-ai-image-quality.md): - In Feb 2026, Google silently swapped the default Gemini app image model from Nano Banana Pro to Nano Banana 2 (Gemini 3.1 Flash, faster + half the cost, ~95% of Pro quality). No announcement. If your AI images suddenly looked flatter o...
- [AI for D2C Brands — The Operator's Playbook — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-17_ronshah_ai-for-d2c-brands-operators-playbook.md): - The AI Opportunity Matrix (2x2): every potential AI app gets scored on (a) impact on unit economics — does this directly improve LTV / reduce CAC / lower COGS, vs nice-to-have, and (b) implementation friction — Zapier integration vs ne...
- [SOP — How to Create Brand Content That Isn't AI Slop](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-17_sop-brand-content-not-ai-slop.md): - AI slop isn't an output problem. It's an input problem. Brands feeding the same publicly available info into the same models expect differentiated results — that's a copy machine pointed at Google's top 10. The brands winning on organi...
- [How to Double Repeat Purchase Rate — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-18_ronshah_double-repeat-purchase-rate.md): - Repeat purchase rate is the real moat in D2C unit economics. Brands at $10M revenue split 25% / 50% repeat rate live in completely different worlds. The first is grinding; the second is compounding.
- [AI Ad Creative Testing Framework — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-19_ronshah_ai-ad-creative-testing-framework.md): - Most D2C brands run 5-10 ads per test, declare a winner, scale. Problem: that's the best of a small sample, not the true best. They're leaving upside on the table.
- [D2C Unit Economics Guide — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-20_ronshah_d2c-unit-economics-guide.md): - The 6 metrics that matter: CAC, LTV, LTV:CAC ratio, contribution margin, CAC payback period, repeat purchase rate (the last covered in sources/2026-03-18ronshahdouble-repeat-purchase-rate).
- [AI Email Marketing for Ecommerce — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-21_ronshah_ai-email-marketing-ecommerce.md): - Email is the highest-ROI channel in D2C (typically 42:1+) and most operators are leaving $500K-$2M on the table by not using AI on it.
- [Scaling a D2C Brand Without an Agency — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-22_ronshah_scaling-d2c-without-agency.md): - Agencies solve for one thing: billable hours. Retainer = no incentive to automate.
- [AI Inventory Planning for D2C — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-23_ronshah_ai-inventory-planning-d2c.md): - Inventory is the easiest way to destroy unit economics. Overstock 20% → tied-up cash → discount spiral. Understock 20% → unrecoverable lost sales. Most operators guess.
- [Powering Up Your Customer Service with Voice AI (Wispr Flow)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-24_guide-customer-service-voice-ai.md): - The thesis: typing forces everyone into flat, corporate-safe tone. Templates strip personality. Volume favors speed over voice. Result: founder voice gets lost as the company scales. Wispr Flow (voice-to-text in any app) restores that...
- [How To Make Money With AI — 19 Ideas (2026)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-24_how-to-make-money-using-ai.md): - Six categories of AI businesses: make/publish, build/automate, market/sell, analyze/model, people/security, operations. 19 ideas total, each with realistic time-to-MVP (1 week to 12 months).
- [D2C Customer Acquisition 2026 — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-24_ronshah_d2c-customer-acquisition-2026.md): - Channel landscape (March 2026): Meta still biggest but flat, TikTok ads cheapest ($15-35 CAC), Google highest ROAS (2.5-5×), email owned channel (35-50:1 ROAS), micro-creator commission model beating mega-influencer flat-fee.
- [Building AI Workflows for Ecommerce — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-25_ronshah_building-ai-workflows-ecommerce.md): - 15+ AI workflows built across operations in the past year. Some game-changers, some interesting, some expensive mistakes. The common thread: workflows that fit the framework win; the others fail.
- [AI Competitive Intelligence for D2C — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-26_ronshah_ai-competitive-intelligence-d2c.md): - Competitive intelligence isn't spying — it's pattern recognition. The brands tracking systematically have a 6-month heads-up on market trends; the ones that don't are always reactive.
- [Claude Cowork Starter Pack](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-31_guide-claude-cowork-starter-pack.md): - Claude Cowork = Claude Code's agentic engine inside a visual desktop app. Same execution capabilities (plan, read files, create documents, connect apps, multi-step tasks) — but no terminal, no coding required. Lives in claude.ai/downlo...
- [I Can't Make a Decision Without AI Anymore — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-03-31_ronshah_cant-make-decision-without-ai.md): - Status report from "further in than most." Ron crossed an invisible line about 6 months ago: AI stopped being a tool he uses and became part of how he thinks.
- [The Prompt That Saved a Media Buyer His Job — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-01_ronshah_prompt-that-saved-media-buyer.md): - Founder ready to fire media buyer over $3,500 ad spend / 3 purchases / $1,100 CPA. Ron asked for 48 hours.
- [Five AI Prompts I Run Every Week — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-02_ronshah_five-ai-prompts-i-run-every-week.md): - Ron's 5 production prompts, run weekly on real data. Plain English, no prompt-engineering tricks. The leverage is in the habit, not the cleverness.
- [A Piece of Paper Just Got Indexed by ChatGPT — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-03_ronshah_paper-just-got-indexed-by-chatgpt.md): - A brand's physical referral pad scored 84% on an AI search readiness assessment. People photograph it in a professional's office → search via ChatGPT → AI extracts text + recommends the brand. Physical-to-digital bridge via phone camer...
- [The $20K Question Every Founder Asks Me — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-04_ronshah_20k-question-every-founder-asks.md): - Most marketing problems aren't budget problems. They're coordination problems. Founder spending $20K/mo across 5 teams (3 agencies + media buyer + freelance creative), flat for 6 months → "you don't have a budget problem, you have a co...
- [What a Retired Amazon VP Taught Me About AI — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-05_ronshah_retired-amazon-vp-taught-me-about-ai.md): - The frame that reorganized Ron's AI use: from a retired Amazon VP over coffee — "Everyone's using AI to do things faster. Almost nobody is using it to see things they couldn't see before. Those are two completely different use cases an...
- [I Gave AI My Entire P&L — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-06_ronshah_i-gave-ai-my-entire-p-and-l.md): - 45-minute AI session on a $20M Google account → found $3M/year in branded-search waste that a sophisticated team with multiple agencies had missed for 2 years.
- [10 Claude Cowork Tips for Ecommerce Teams](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-07_10-claude-cowork-tips-ecommerce-teams.md): - Power tips that separate "tried it once" from "this is now part of how I operate" — assumes baseline setup from sources/2026-03-31guide-claude-cowork-starter-pack.
- [The Gap Between AI Users and AI Thinkers — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-07_ronshah_gap-between-ai-users-and-ai-thinkers.md): - Two types of AI use, separated by a psychological line:
- [My Team Taught Me an AI Trick That Generates Better Ads — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-08_ronshah_team-taught-me-ai-trick-better-ads.md): - Team member showed Ron a new prompt: "Analyze these 10 ads. What emotional trigger is present in every single winner that is absent from every loser? Don't tell me about format or structure. Tell me about the feeling."
- [How I Use AI to Run 5 Businesses Without Losing My Mind — Ron Shah](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-09_ronshah_run-5-businesses-without-losing-mind.md): - Ron runs 5 businesses (advisory clients, agency, media co, health venture). 2 years ago: barely managing, reactive, fires-only. AI changed it structurally, not incrementally.
- [Amazon Search Query Performance Analysis with Claude Cowork](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-14_guide-amazon-search-query-performance-claude-cowork.md): - Concrete instance of sources/2026-04-0710-claude-cowork-tips-ecommerce-teams tip 3: a custom Cowork Skill turns Amazon's dense Search Query Performance (SQP) report into a structured 6-layer analysis. Without the skill, Claude gives a...
- [AI Knowledge Layer — Why Your Agents Are Useless Without It (Shann)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-15_ai-knowledge-layer-shann.md): - The AI knowledge layer is the infrastructure that sits between you and your agents — what they read before doing anything. Without it agents guess; with it they know your voice, data, patterns. Two layers: Knowledge Base Layer (KBL, dy...
- [Scale Your Product Images with JSON-Based Prompting](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-21_guide-scale-product-images-json-prompting.md): - The shift: from creating images from a prompt to creating images from a system. JSON converts an established image into a structured, editable recipe — change one field (background, props, framing) without rebuilding the rest.
- [Build Marketing Assets in Minutes with Claude Design](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-28_guide-build-marketing-assets-claude-design.md): - Claude Design = Anthropic's visual creation tool, launched April 2026 as a research preview at claude.ai/design. Powered by Claude Opus 4.7. You describe what you want in plain language → Claude asks clarifying questions → generates a...
- [Agentic Commerce on Shopify — How It Works (2026)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-28_how-agentic-commerce-works.md): - AI-driven traffic to Shopify stores grew 8× YoY; AI-attributed orders grew 15× YoY since January 2025.
- [Get RICH in the AI Revolution (2026) — Alex Hormozi](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-30_hormozi_get-rich-in-the-ai-revolution-2026.md): - Hormozi's 3-pillar AI system (parallel to our topics/ai-knowledge-layer architecture but operator-framed): Business Context (your business write-up) + SOP / Prompt Repository (prompts as SOPs 2.0 for repeated work) + Data Sources (expo...
- [How to Use AI in Your Business in 2026 — Alex Hormozi](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-30_hormozi_how-to-use-ai-in-your-business-in-2026.md): - You don't have to become an AI business. You need to USE AI in your business. Same as the internet — almost every business is on the internet, but few are "internet businesses." Hormozi's central reframe.
- [How to Win With AI in 2026 — Alex Hormozi](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04-30_hormozi_how-to-win-with-ai-in-2026.md): - Stop role-based thinking. Start workflow-based thinking. For every hire, list 4-10 things they actually do; ask if each could be a workflow instead of headcount. Manufacturing-style linear flow human-organized hierarchy.
- [14 Claude-for-Ecom Tactical Tips (X.com, Q1-Q2 2026)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026-04_x-tips-claude-for-ecom.md): - 14 short case studies of operators using Claude on raw ecom data (Shopify exports, Meta ads, support tickets, returns, 3PL invoices, subscription cancellations) to find non-obvious problems and reframe strategy.
- [AI Inside Enterprises & How Your Brand Can Beat Them](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_ai-inside-enterprises.md): - "95% of AI pilots fail." Craig confirms via Crocs experience: 200+ AI startups talked to, 40-50 pilots launched. The reasons for failure: integration overhead, weak adoption, no executive champion, "tools that became features in ChatGP...
- [Brand Is Something AI "Can't Replicate," Or Can It?](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_brand-is-something-ai-cant-replicate.md): - Counter-thesis on AI creative (Taylor Holiday): A study (Eric Seaffort, Mobile Dev Memo) compared human designers vs human+AI vs unchained AI (no constraints) — the unchained AI won. Reason: humans can only extrapolate from their lived...
- [The First Ecommerce Jobs AI Wipes Out & What Comes Next](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_first-ecom-jobs-ai-wipes-out.md): - "SAAS is a sinking ship" — Amit Pasha (Rich Panel founder) admits his own business model is dying. Per Dario Amodei (Anthropic CEO): Phase 1 = 90% of code by AI. Phase 2 = 100% of code by AI (where we are now). Phase 3 (next 12 months)...
- [How to Actually Start Using AI Agents (Solo OpenClaw Setup)](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_how-to-actually-start-using-ai-agents.md): - A solo episode where Nik Sharma walks through setting up an OpenClaw / Moltbot agent on a Mac mini step-by-step, in real time, with the listener following along. Pairs with sources/2026podcasts15e10why-ai-is-the-next-industrial-revolut...
- [Performance Creative — Velocity, AI Tooling, and the Paid-Organic Relationship](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_performance-creative-velocity-ai-tooling.md): - 2026 = "operations and velocity" (Dar's yearly buzzword timeline). 2023 was creative strategy. 2024 was diversity. 2025 was volume. 2026: how quickly can you get great ideas and content into the ad account?
- [Rethinking SMS Marketing for Ecommerce — AI, RCS & LTV](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_rethinking-sms-marketing-ai-rcs-ltv.md): - 8-10 messages per subscriber per month = sweet spot. Most brands send 2 (way too few). Some 8-9-figure brands send 23 in first 30 days (Grünes / Grunes). "If you annoyed people it would show up as unsub rate. Send more until you see it...
- [S14 E5 — How AI Is Rewriting the Rules of Ecom Marketing](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_s14e5-how-ai-rewriting-rules-ecom-marketing-liam-millward.md): - Liam's stage line: "If AI isn't driving at least 25% of your revenue today across email, SMS, retention — you're already failing."
- [S15 E10 — Why AI Is the Next Industrial Revolution](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_s15e10_why-ai-is-the-next-industrial-revolution.md): - The 2-week shift. Up until ~Feb 2026 most operators "knew AI was here, planned to incorporate it." In 2 weeks (per Sharma's recording around March 2026), the gap between users and non-users widened sharply. New models drop every 2-3 we...
- [From Spreadsheets to AI Agents — The Ecommerce Data Playbook](https://midas-wiki.vercel.app/kb/ai-ecom/sources/2026_podcast_spreadsheets-to-ai-agents.md): - Without a data foundation, AI is useless. The thesis of the episode: clean unified data warehouse is the prerequisite for any meaningful agentic data layer. Saras Analytics' core business for 10 years (data warehouse setup) becomes 100...
### Tools
- [AI Content Detectors](https://midas-wiki.vercel.app/kb/ai-ecom/tools/ai-content-detectors.md): Software tools that classify text as AI-generated vs human-written. All five major tools work on the same two signals: burstiness and perplexity. Differences are in pricing, additional features (plagiarism checking, multi-modal detection...
- [Claude Cowork](https://midas-wiki.vercel.app/kb/ai-ecom/tools/claude-cowork.md): Anthropic's agentic desktop workspace — Claude Code's execution engine wrapped in a visual interface anyone can use. Same capabilities as Claude Code (plan, read files, create documents, connect to apps, multi-step tasks) but no terminal...
- [Claude Design](https://midas-wiki.vercel.app/kb/ai-ecom/tools/claude-design.md): Anthropic's visual creation tool, launched April 2026 as a research preview. Lives at claude.ai/design. Powered by Claude Opus 4.7 (Anthropic's most capable vision model). You describe what you want in plain language → Claude asks clarif...
- [Claude](https://midas-wiki.vercel.app/kb/ai-ecom/tools/claude.md): Anthropic's LLM family. The most-cited single tool across the wiki so far. Used both as a generic LLM (data analysis, content) and as the engine for Claude Code — the CLI that operates on a folder of files (the schema for this wiki, Shan...
- [Gemini](https://midas-wiki.vercel.app/kb/ai-ecom/tools/gemini.md): Google's AI family. By Q2 2026, the dominant tool for ecom creative work (image generation via Nano Banana 2 / Pro), multimodal analysis (UX audit with screenshots), and video transcription (TikTok pattern analysis).
- [Genspark](https://midas-wiki.vercel.app/kb/ai-ecom/tools/genspark.md): AI tool (genspark.ai) with a Download Agent feature that collects videos at scale from URLs. In the wiki, cited specifically for its role in the TikTok pattern-mining pipeline — bulk-downloading competitor TikToks for analysis without ma...
- [GPT for Sheets / Excel](https://midas-wiki.vercel.app/kb/ai-ecom/tools/gpt-for-sheets-excel.md): Talarian's "GPT for Work" add-ons embed GPT inside Google Sheets and Microsoft Excel. The point: bulk text processing without leaving the spreadsheet. Eliminates export-to-external-tool friction and version control issues.
- [NotebookLM](https://midas-wiki.vercel.app/kb/ai-ecom/tools/notebooklm.md): Google's lightweight knowledge-base tool. Free, browser-based, queryable. Best understood as a read-only knowledge layer: you upload sources, query them with citations, but NotebookLM doesn't modify or create files in your local environm...
- [Obsidian](https://midas-wiki.vercel.app/kb/ai-ecom/tools/obsidian.md): Local-first markdown editor used as the front-end IDE for the knowledge layer. Pairs with Claude Code (the editor / agent) the way an IDE pairs with a programmer. The wiki is a folder of .md files; Obsidian renders, searches, and graph-v...
- [Wispr Flow](https://midas-wiki.vercel.app/kb/ai-ecom/tools/wispr-flow.md): Voice-to-text tool that works inside any app — helpdesk, email, Slack, Amazon Seller Central. Turns natural speech into clean, formatted text. Auto-tone-matches to the app context (more formal in email, conversational in live chat). Remo...
### Topics
- [Agentic Commerce](https://midas-wiki.vercel.app/kb/ai-ecom/topics/agentic-commerce.md): The shift from storefront-mediated to agent-mediated discovery and purchase. Shoppers interact with AI agents (ChatGPT, Copilot, Gemini, Google AI Mode) to find, compare, and buy — instead of browsing brand sites or Google search results...
- [AI Amplification, Not Automation](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-amplification-not-automation.md): A reframe of how to think about what AI does in a working business. Most operator-facing AI marketing sells automation (infinite leverage, runs forever, never breaks). Most operator reality is amplification (60-90 min task → 15-20 min ta...
- [AI Business Models](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-business-models.md): Catalog of viable AI business categories and their realistic timelines. Useful when thinking about service offerings, productized solutions, or side projects. Source: Shopify's 2026-03 "19 ideas" guide (sources/2026-03-24how-to-make-mone...
- [AI Content Creation Workflows](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-content-creation-workflows.md): Specific workflows where AI substantially compresses the time/cost of producing a polished output. Each is documented as a standalone workflow page; this topic page is the navigation hub that ties them together and surfaces the cross-cut...
- [AI Content Detection](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-content-detection.md): A narrow but real operating concern: tools that classify text as AI-generated vs human-written. Useful for QA on outsourced content, fake reviews, and screening AI-generated job applications. Not useful as a publishing gate — too many fa...
- [AI Creative Pipeline](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-creative-pipeline.md): The end-to-end stack for producing brand creative — product images, infographics, decks, landing pages — with AI. By Q2 2026 this is no longer a single-tool conversation; it's a layered pipeline where different tools handle different sta...
- [AI Customer Service](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-customer-service.md): The least-mature application of AI in ecom relative to the consumer expectation. Pure-AI CS is a regression: 87% of consumers prefer hybrid (human + AI) support over pure AI (topics/ai-market-data). The winning operators use AI to amplif...
- [AI Knowledge Layer](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-knowledge-layer.md): The infrastructure that sits between operators and their AI agents. The thing the agent reads before doing anything. Without it, agents guess. With it, they know your voice, data, patterns.
- [AI Market Data](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-market-data.md): Living catalog of market-size, adoption, sentiment, and behavior datapoints across the AI / ecom intersection. Updated on every ingest. Numbers are kept with their source date so we can see velocity.
- [AI Trends 2026](https://midas-wiki.vercel.app/kb/ai-ecom/topics/ai-trends-2026.md): Macro-level trends shaping ecom and business operations in 2026. The list below is the synthesis from Shopify's 2025-07 trends piece, with cross-references where other sources reinforce or contradict.
- [Amazon Ecom with AI](https://midas-wiki.vercel.app/kb/ai-ecom/topics/amazon-ecom-with-ai.md): Amazon-specific application of AI tools and patterns. Three categories of work where AI delivers the largest 2026 leverage: listing images (Nano Banana Pro pipeline), listing copy (Rufus-ready, customer-language-driven, anti-slop), perfo...
- [Brand Voice vs AI Slop](https://midas-wiki.vercel.app/kb/ai-ecom/topics/brand-voice-vs-ai-slop.md): The single largest content-quality problem in AI-augmented ecom: every brand using the same models gets the same default voice. The brands that maintain a distinctive voice while using AI win. The brands that don't dissolve into the noise.
- [Claude for Ecom Data Analysis](https://midas-wiki.vercel.app/kb/ai-ecom/topics/claude-for-ecom-data-analysis.md): A specific high-leverage application of Claude: feeding raw ecom data (already collected, already accessible) into Claude with a single sharp question. The reusable patterns sit at the intersection of operator instinct and synthesis band...
- [D2C Unit Economics — Benchmarks + Diagnostic Playbook](https://midas-wiki.vercel.app/kb/ai-ecom/topics/d2c-unit-economics.md): The neutral economic framework underneath every AI workflow in this wiki. AI doesn't fix bad unit economics — it amplifies whatever is already there. So before applying any AI lever, an operator needs to know which metric is actually bro...
- [Team Knowledge Systems](https://midas-wiki.vercel.app/kb/ai-ecom/topics/team-knowledge-systems.md): The application of topics/ai-knowledge-layer thinking to teams of 1-50 people: how AI-readable knowledge bases plus role-tuned agents turn scattered information into queryable infrastructure. By 2026, this is the dominant operating patte...
### Workflows
- [AI Creative Testing — 25 Variations + 3-Bucket Framework](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/ai-creative-testing-25-variations.md): - Brand running 5-10 creatives per test → seeing the "best of small sample" problem.
- [AI Image Creation Stack](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/ai-image-creation-stack.md): The consolidated end-to-end view of producing AI-generated brand and product imagery in 2026. Combines four sub-workflows into a coherent stack: capture → generate → quality-route → scale.
- [AI Inventory Planning for D2C — Demand Forecasting Stack](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/ai-inventory-planning-d2c.md): - Brand has 18-24 months of historical sales data + experiences seasonal demand swings.
- [AI-Powered UX Audit](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/ai-powered-ux-audit.md): A multi-site UX audit (your site + 3 competitors) producing a competitive matrix + SWOT + Top 3 friction killers + 90-day roadmap — in a single Gemini 3 session. What used to be a thousands-of-dollars, weeks-long agency engagement now fi...
- [AI Search Readiness Audit (GEO for D2C)](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/ai-search-readiness-audit.md): - Brand discovers it's not appearing in ChatGPT / Perplexity / Gemini results for category queries.
- [Amazon Product Images from Phone Photos](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/amazon-product-images-from-phone-photos.md): Convert phone photos into Amazon-listing-ready images in ~90 minutes. Replaces the 3-week wait for product photography. Sub-workflow of the broader workflows/ai-image-creation-stack.
- [Amazon SQP Deep Analysis with Claude Cowork](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/amazon-sqp-analysis-with-cowork.md): Turn Amazon's dense Search Query Performance report into a structured 6-layer analysis: brand health + funnel diagnostic + competitive intelligence + keyword strategy refresh — in a single Cowork session. The pattern is a concrete instan...
- [Anti-Slop Brand Content — The 3-Asset Foundation + Production SOP](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/anti-slop-brand-content-sop.md): The canonical workflow for using AI to produce content that sounds like a specific brand instead of like generic AI. This is the most important workflow in the wiki for content-heavy ecom operators. AI slop isn't an output problem. It's...
- [Brand-Consistent Infographics with Gemini](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/brand-consistent-infographics.md): Convert blog posts, reports, or text into polished brand-consistent infographics in ~15 minutes. The key insight: a structured 6-attribute style guide + a two-step prompt that separates text parsing from style application defeats Nano Ba...
- [Build Marketing Assets in Minutes with Claude Design](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/claude-design-marketing-assets.md): End-to-end workflow for using Claude Design (claude.ai/design, powered by Opus 4.7) to produce campaign landing pages, pitch decks, animated videos, email layouts, and one-pagers in minutes instead of days.
- [Claude × Ecom Data — 14 Reusable Analysis Patterns](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/claude-ecom-data-analysis-patterns.md): Operator-tested patterns from real ecom brands for using Claude as a data analyst. Each pattern includes: situation, data inputs, prompt template, expected output, and an action shape. All 14 documented in sources/2026-04x-tips-claude-fo...
- [Claude in PowerPoint — Template-Aware Deck Generation](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/claude-in-powerpoint.md): Claude as official Microsoft add-in inside PowerPoint. Template-aware — reads slide masters, layouts, fonts, colors. Generate and reshape slides without leaving the app.
- [Competitive Intelligence Automation for D2C](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/competitive-intelligence-automation.md): - Brand is consistently 4-6 months behind competitor moves (you spot the trend after they own the channel).
- [Claude Cowork — Setup + 10 Power Tips for Ecom Teams](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/cowork-setup-and-power-tips.md): End-to-end guide to going from "tried it once" to "this is now part of how I operate." Combines the starter-pack setup (sources/2026-03-31guide-claude-cowork-starter-pack) with the 10 power tips (sources/2026-04-0710-claude-cowork-tips-e...
- [Diagnose + Fix AI Image Quality](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/diagnose-fix-ai-image-quality.md): Practical troubleshooting workflow for sudden quality drops in AI image generation, plus the model-routing strategy that prevents most issues. Sub-workflow of workflows/ai-image-creation-stack.
- [AI Email Marketing — 4-Lever Playbook for D2C](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/email-marketing-ai-levers.md): - Brand at $1M-$30M ARR with email contributing 30-60% of revenue.
- [Excel Bulk Text Processing with AI](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/excel-bulk-text-with-ai.md): Bulk text processing inside Google Sheets and Microsoft Excel via Talarian's GPT for Work add-on. Eliminates export-to-external-tool friction and version control issues. Three production-ready apps: SEO metadata, localization, sentiment...
- [Five AI Prompts to Run Every Week (Ron Shah's Canon)](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/five-ai-prompts-i-run-every-week.md): - A new client wants the fastest, cheapest, highest-leverage AI rollout — these 5 prompts ship value in week 1 with no infrastructure.
- [Full-Funnel Skeptic Audit (Ron Shah's Diagnostic Prompt)](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/full-funnel-skeptic-audit.md): - ROAS / CPA is bad but you can't tell which layer is broken (ads? LP? quiz? offer?).
- [Hormozi's 3-Pillar AI System](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/hormozi-3-pillar-ai-system.md): End-to-end setup workflow for Alex Hormozi's operator-framed knowledge-layer system. Same architecture as our topics/ai-knowledge-layer but with operator vocabulary, calendar-block tactics, and prompt-chain patterns. Use this as the entr...
- [Humanize AI-Generated Content](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/humanize-ai-content.md): A repeatable pass to convert default-AI text into content that sounds like a specific brand. Use after any AI-assisted draft, before publishing.
- [JSON-Based Image Prompting](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/json-image-prompting.md): Move from creating images from a prompt to creating images from a system. JSON converts an established image into a structured, editable recipe — change one field (background, props, framing) without rebuilding the rest. Unlocks scale ac...
- [NotebookLM Client Brain — Agency SOP](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/notebooklm-client-brain.md): Per-client knowledge base built in NotebookLM. Centralizes scattered client information into a queryable source. Preps calls in <5 minutes, onboards new team members in hours instead of weeks, grounds every strategy decision in actual cl...
- [Doubling Repeat Purchase Rate — 4-Lever Playbook](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/repeat-purchase-rate-doubling.md): - Brand at $3M-$30M ARR with repeat purchase rate <30% (90-day window).
- [Set Up an AI Knowledge Layer](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/setup-knowledge-layer.md): End-to-end workflow for setting up the knowledge layer pattern from zero. Adapted from Shann Holmberg's llm-wikid (sources/2026-04-15ai-knowledge-layer-shann). 20-minute first pass, but the value compounds over weeks.
- [Viral TikTok Script Machine](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/tiktok-script-machine.md): A 4-step pipeline that converts competitor TikTok analysis into a compounding script-generation engine. Genspark downloads videos at scale → Gemini transcribes + analyzes each → pattern library accumulates → Claude generates new scripts...
- [Voice-AI Customer Service with Wispr Flow](https://midas-wiki.vercel.app/kb/ai-ecom/workflows/voice-cs-with-wispr-flow.md): Make every CS reply sound like the founder wrote it — at scale. Voice-to-text removes typing as the bottleneck while preserving founder voice. 15-30 seconds per ticket vs 3-5 minutes typing from scratch.

## Building with Claude Code

Patterns for building deliverables with Claude Code and coding agents: workflows, prompting, subagents, skills, and shipping.

Full bundle: [claude-code.txt](https://midas-wiki.vercel.app/kb/_full/claude-code.txt)

- [Wiki Index](https://midas-wiki.vercel.app/kb/claude-code/index.md): Content catalog. Every page listed with link + one-line summary. Updated on every ingest.
- [Log](https://midas-wiki.vercel.app/kb/claude-code/log.md): Append-only chronological record. Format: [YYYY-MM-DD] ingest|query|lint | <title
- [How MIDAS Builds with Claude Code in 2026](https://midas-wiki.vercel.app/kb/claude-code/overview.md): Evolving thesis. Built from ingesting 27 source files (Anthropic official docs + Boris Cherny's 89-tip site + Walking Labs harness-engineering curriculum + Karpathy autoresearch).
### People
- [Anthropic Claude Code Team](https://midas-wiki.vercel.app/kb/claude-code/people/anthropic-claude-code-team.md): The team at Anthropic responsible for building, evolving, and documenting Claude Code. Includes Boris Cherny (lead creator, see boris-cherny once that page exists) and a broader team referenced collectively in official docs.
- [Boris Cherny](https://midas-wiki.vercel.app/kb/claude-code/people/boris-cherny.md): Lead creator of Claude Code at Anthropic. Author of 89+ practical tips on howborisusesclaudecode.com (a SPA built entirely with Claude Code, by Boris's collaborator).
- [Andrej Karpathy](https://midas-wiki.vercel.app/kb/claude-code/people/karpathy.md): Former Director of AI at Tesla, founding member of OpenAI, prolific educator (nanochat, nanoGPT, Zero to Hero series). Built autoresearch as a public demonstration of the autonomous AI-driven research pattern.
- [Walking Labs](https://midas-wiki.vercel.app/kb/claude-code/people/walking-labs-authors.md): Authors of the "Learn Harness Engineering" curriculum at walkinglabs.github.io. Identity not specified on the public-facing site; treated as an org/team contribution rather than an individual.
### Playbooks
- [Playbook: Bootstrap nowego klient-repo w Claude Code](https://midas-wiki.vercel.app/kb/claude-code/playbooks/midas-client-repo-bootstrap-pl.md): Nowy projekt klienta MIDAS. Zaraz po kickoff call, zanim zaczniesz pisać kod. Inwestujesz 30-45 minut raz, zwraca się to dziesiątkami godzin przez czas trwania projektu.
- [Playbook: Pre-delivery verification gate (MIDAS)](https://midas-wiki.vercel.app/kb/claude-code/playbooks/midas-pre-delivery-verification-pl.md): Zawsze przed wysłaniem czegokolwiek do klienta MIDAS. Nigdy nie shippuj bez przejścia tego checklistu.
- [Playbook: Source-attribution prompting dla MIDAS deliverable'ów](https://midas-wiki.vercel.app/kb/claude-code/playbooks/midas-source-attribution-pl.md): Każdy LLM-generated artefakt który ma zostać:
### Sources
- [Source: Anthropic — Best practices for Claude Code](https://midas-wiki.vercel.app/kb/claude-code/sources/anthropic-best-practices.md): - Context window is the master constraint. Performance degrades as context fills. Everything else flows from this.
- [Source: Boris's 12 Parts — Feature Catalog](https://midas-wiki.vercel.app/kb/claude-code/sources/boris-parts-catalog.md): This source is a catalog, not a synthesis. The synthesis lives in:
- [Source: Boris's Brain — Curated Playbook](https://midas-wiki.vercel.app/kb/claude-code/sources/boris-playbook-curated.md): - 18 entries auto-distilled from 20 of Boris's 87 tips via Anthropic Dreaming research preview.
- [Source: Dreaming Meta-Experiment + Feedback Log](https://midas-wiki.vercel.app/kb/claude-code/sources/dreaming-meta-experiment.md): - The author ran Anthropic's Dreaming research preview over 20 of Boris's tips. First-pass output was polished but full of fabrications (invented industry trends, made-up stats, simulated examples promoted to "Observations").
- [Source: Karpathy — autoresearch](https://midas-wiki.vercel.app/kb/claude-code/sources/karpathy-autoresearch.md): - Pattern: give an AI agent a closed-loop optimization problem (modify code → run for fixed time → measure single metric → keep/discard → loop), let it iterate autonomously while you sleep
- [Source: Walking Labs — Learn Harness Engineering](https://midas-wiki.vercel.app/kb/claude-code/sources/walkinglabs-harness-engineering.md): - Thesis: agent reliability is bottlenecked by the execution environment (the "harness"), not by model capability. Upgrading models won't fix what bad harnesses break.
- [Source: Zero to Claude Code (zero2claude.dev)](https://midas-wiki.vercel.app/kb/claude-code/sources/zero2claude.md): - 148-lesson interactive web course (14 levels, free, by Itay Shmool, 22K+ students). Anchored on Anthropic best-practices but pedagogically rebuilt for absolute beginners with virtual terminal, fill-in-blank, drag-match, quiz, step-thro...
### Tools
- [Agent View and `/goal`](https://midas-wiki.vercel.app/kb/claude-code/tools/agent-view-and-goal.md): Two newest features (May 2026) that change how you manage multi-agent and long-running work.
- [AGENTS.md](https://midas-wiki.vercel.app/kb/claude-code/tools/agents-md.md): The framework-agnostic equivalent of CLAUDE.md. Used by OpenAI Codex / Codex-CLI and other coding agents that follow the same "rules file in the repo root" pattern.
- [Auto Mode](https://midas-wiki.vercel.app/kb/claude-code/tools/auto-mode.md): Permission mode where a separate classifier model reviews commands before they run. Blocks risky actions (scope escalation, unknown infrastructure, hostile content), lets routine work proceed without prompts.
- [Checkpoints and Rewind](https://midas-wiki.vercel.app/kb/claude-code/tools/checkpoints-rewind.md): Every prompt you send creates a checkpoint. Claude automatically snapshots files before each change. You can restore conversation, code, or both to any previous checkpoint.
- [CLAUDE.md](https://midas-wiki.vercel.app/kb/claude-code/tools/claude-md.md): A markdown file Claude Code reads at the start of every conversation. Persistent context that Claude can't infer from code alone.
- [CLI Tools](https://midas-wiki.vercel.app/kb/claude-code/tools/cli-tools.md): External command-line tools Claude Code uses via Bash: gh (GitHub), gcloud, aws, sentry-cli, stripe, vercel, etc.
- [Hooks](https://midas-wiki.vercel.app/kb/claude-code/tools/hooks.md): Scripts that run automatically at specific points in Claude's workflow. Unlike CLAUDE.md instructions which are advisory, hooks are deterministic — they guarantee the action happens.
- [`/loop` and `/schedule` (Routines)](https://midas-wiki.vercel.app/kb/claude-code/tools/loop-and-schedule.md): Two related primitives for running Claude on a recurring or event-driven basis without you in the loop.
- [MCP Servers](https://midas-wiki.vercel.app/kb/claude-code/tools/mcp-servers.md): Model Context Protocol servers. External integrations Claude Code can call as tools. Examples: Notion, Figma, your database, GitHub, Slack, Asana, Gmail.
- [Model and Effort Selection](https://midas-wiki.vercel.app/kb/claude-code/tools/model-selection.md): /model switches the Claude model and effort level in Claude Code. Default behavior matters more than people think — it affects every task, not just hard ones.
- [Non-Interactive Mode (`claude -p`)](https://midas-wiki.vercel.app/kb/claude-code/tools/non-interactive-mode.md): Run Claude Code without an interactive session: one-shot prompt, parse the output programmatically.
- [Permissions](https://midas-wiki.vercel.app/kb/claude-code/tools/permissions.md): Claude Code requests permission for actions that might modify your system: file writes, Bash commands, MCP tool calls. By default, this is safe but tedious — after the 10th approval, you're not really reviewing, just clicking.
- [Plan Mode](https://midas-wiki.vercel.app/kb/claude-code/tools/plan-mode.md): A permission mode where Claude can read files and answer questions but cannot make changes. Used to separate exploration/planning from execution.
- [Plugins](https://midas-wiki.vercel.app/kb/claude-code/tools/plugins.md): Bundled units of skills, hooks, subagents, MCP servers. Installable from a community/Anthropic marketplace.
- [Recaps, Focus Mode, `/go`](https://midas-wiki.vercel.app/kb/claude-code/tools/recaps-focus-go.md): Three Opus 4.7-era quality-of-life features that change how you interact with Claude Code during long tasks.
- [`/simplify` and `/batch`](https://midas-wiki.vercel.app/kb/claude-code/tools/simplify-and-batch.md): Built-in skill that reviews changed code for reuse, quality, efficiency, then fixes issues. Run before PR to catch:
- [Skills](https://midas-wiki.vercel.app/kb/claude-code/tools/skills.md): Folder + SKILL.md file in .claude/skills/<name/. Domain knowledge or reusable workflow that Claude loads on demand, unlike CLAUDE.md which loads every session.
- [Slash Commands](https://midas-wiki.vercel.app/kb/claude-code/tools/slash-commands.md): Markdown files in .claude/commands/ that define repeatable workflows you invoke with /<command-name. Like skills, but explicitly callable rather than auto-applied.
- [Subagents](https://midas-wiki.vercel.app/kb/claude-code/tools/subagents.md): Specialized assistant defined in .claude/agents/<name.md. Runs in its own context window with its own set of allowed tools. The parent Claude delegates a task to it; the subagent works in isolation and returns a summary.
- [Worktrees](https://midas-wiki.vercel.app/kb/claude-code/tools/worktrees.md): Native git worktree support inside Claude Code. Each session runs in an isolated git checkout so parallel Claude sessions don't collide on edits.
### Topics
- [Autonomous-Loop Pattern](https://midas-wiki.vercel.app/kb/claude-code/topics/autonomous-loop-pattern.md): Any closed-loop optimization problem with (1) a mutable artifact, (2) a deterministic single-number eval, and (3) a bounded per-iteration cost can be handed to Claude to run autonomously while you sleep. The discipline that makes this wo...
- [Compounding Engineering](https://midas-wiki.vercel.app/kb/claude-code/topics/compounding-engineering.md): [Boris] Every correction you make to Claude should permanently update the rules that govern future sessions. Done consistently, the rules file becomes a compounding asset — each session's quality improves over time because past mistakes...
- [Context Window Management](https://midas-wiki.vercel.app/kb/claude-code/topics/context-window-management.md): The context window is the master constraint of Claude Code performance. Every other best practice traces back to one of two failure modes: (1) context fills too fast, or (2) context contains noise that drowns out signal. Treat context as...
- [Common Failure Patterns](https://midas-wiki.vercel.app/kb/claude-code/topics/failure-patterns.md): Most Claude Code failures fit into five recognizable patterns. Naming them lets you spot the failure mode early and apply the right fix instead of grinding.
- [Harness Engineering](https://midas-wiki.vercel.app/kb/claude-code/topics/harness-engineering.md): Agent reliability is bottlenecked by the execution environment (the harness), not by the underlying model. Upgrading models won't fix problems that bad harnesses create. The discipline of designing this environment — rules files, runtime...
- [Parallelism](https://midas-wiki.vercel.app/kb/claude-code/topics/parallelism.md): Single Claude, single conversation, single workstream is the default — but it's not the ceiling. Claude Code scales horizontally via worktrees, parallel sessions, fan-out scripting, and agent teams. The multiplier comes from: (1) isolati...
- [Session Management](https://midas-wiki.vercel.app/kb/claude-code/topics/session-management.md): Sessions are persistent and reversible. Treat them like git branches: nameable, resumable, forkable. The cost of resetting is near-zero; the cost of dragging stale context is high. Course-correct early; restart often.
- [Source-Attribution Prompting](https://midas-wiki.vercel.app/kb/claude-code/topics/source-attribution-prompting.md): When an LLM generates a synthesis (review, summary, report, proposal), the default output mixes verified claims with plausible-sounding extrapolations and invented examples. The reader cannot tell which is which. Adding a structural prom...
- [Prompt Specificity](https://midas-wiki.vercel.app/kb/claude-code/topics/specificity.md): Claude can infer intent but cannot read your mind. The more precise your prompt, the fewer corrections you need. Specificity costs you 30 seconds of writing; ambiguity costs you 30 minutes of back-and-forth.
- [Verification](https://midas-wiki.vercel.app/kb/claude-code/topics/verification.md): Verification is the single highest-leverage practice in Claude Code. Without clear success criteria, Claude produces plausible-looking work that doesn't actually function. You become the only feedback loop, and every mistake requires you...
### Workflows
- [Workflow: Bug Fixing via Raw Artifact](https://midas-wiki.vercel.app/kb/claude-code/workflows/bug-fixing-via-raw-artifact.md): The bug already lives as a raw artifact:
- [Workflow: Experiment Loop with Budget (autoresearch-style)](https://midas-wiki.vercel.app/kb/claude-code/workflows/experiment-loop-with-budget.md): You have:
- [Workflow: Explore → Plan → Implement → Commit](https://midas-wiki.vercel.app/kb/claude-code/workflows/explore-plan-implement-commit.md): Any task where:
- [Workflow: Fan Out Across Files](https://midas-wiki.vercel.app/kb/claude-code/workflows/fan-out-across-files.md): - Large migrations (every file in a directory needs the same transformation)
- [Workflow: Interview-Then-Spec](https://midas-wiki.vercel.app/kb/claude-code/workflows/interview-then-spec.md): - Larger features where you haven't thought through all the edge cases
- [Workflow: Prove-It Prompting](https://midas-wiki.vercel.app/kb/claude-code/workflows/prove-it-prompting.md): After Claude produces a solution that looks right but you want to:
- [Workflow: Role Separation (Planner / Generator / Evaluator)](https://midas-wiki.vercel.app/kb/claude-code/workflows/role-separation-pattern.md): - High-stakes implementations where "satisfied with its own work" failure mode is unacceptable
- [Workflow: Writer / Reviewer Pattern](https://midas-wiki.vercel.app/kb/claude-code/workflows/writer-reviewer-pattern.md): - Any non-trivial implementation you want to ship to a client