Image: GitHub
Why it mattersMost AI-app guides list tools and stop. This one gives the price per 1,000 users for each choice. A founder can read it in a morning.
Blake Anderson has posted a 14-chapter playbook for building profitable AI apps. He says he scaled five apps to over $100,000 in monthly revenue in the past three years. He also says the first version of EcoGPT was built on this stack. The repository went up on 26 October. It had 569 stars three days later.
The playbook is a Markdown repo. The reader clones it and points a coding agent at the top-level AGENTS.md file. The agent reads the chapters as it works.
The stack it names
The playbook picks one tool for each job. Blake Anderson says the backend is Cloudflare. He names Workers for the API, D1 for the database, R2 for files. He also names KV, Durable Objects, Queues, Workflows and the AI Gateway. Login is Better Auth. For traffic that outgrows D1, he suggests PlanetScale.
For the model, he picks DeepSeek through OpenRouter as the default. He writes that teams running one AI workflow at scale save money on Vast.ai or RunPod. Payments are Superwall on mobile and Stripe on the web. Analytics and errors share one pipeline on Tinybird.
The clients are SwiftUI for iOS. React Native or Kotlin for Android. React with Vite for the web. The coding agent is Claude Code, or OpenCode with DeepSeek.
What not to use
The README has a short line that will draw argument. It names the tools he says to avoid: Vercel, Netlify, Lovable, Bolt, v0, Replit hosting, Bubble, FlutterFlow, Firebase, Supabase, Clerk, Auth0, Sentry, PostHog, Next.js and ORMs. No reason is given on the top page. The costs chapter has side-by-side prices for each one.
The prices
The costs chapter gives list prices for every tool in the stack. Blake Anderson says he read these prices on 26 October 2026. He also lists the "popular" tool each one replaces.
His benchmark funnel is simple. Thirty percent of downloads start a trial. Thirty percent of trials pay $30 a year. That works out to $2.70 of revenue per download.
At 100,000 downloads a month, the stack costs about $690. Superwall takes $2,700 at one percent. Cloudflare takes $190. Tinybird takes $275. The AI calls take $3,500 on DeepSeek V4.1 Flash. The AI number is 60 calls per user per month.
He gives the cost per 1,000 users of each tool against the popular one it replaces. Cloudflare Workers at $0.21 against Vercel Pro at $3.48. Better Auth at $0 against Clerk at $10.25. DeepSeek at $29.43 against OpenAI gpt-6.1-sol at $420. These are Blake Anderson's own numbers, from his own usage model.
The 14 chapters
Phase 0 is picking the niche. The scorecard rejects any idea that scores below 25 out of 35. Phases 1 to 12 are the build: brief, repo, backend, auth, AI, web, iOS and Android, payments, analytics, onboarding, retention and launch. Phase 13 is the first 1,000 users. Phase 14 is one funnel test and one product fix every week.
Every chapter has a "done when" list. The agent working through the playbook knows when a step is finished. Templates in the repo include a worker.ts, an auth.ts, a wrangler.jsonc, an AGENTS.app.md and a dev.vars.example. The repo is MIT licensed.
Source
ai-app-playbook by Blake Anderson on GitHub.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.