Point an agent at your funnel. Wake up to the winner.
Karpathy open-sourced the methodology (88,000+ GitHub stars, March 2026). We turned it into 12 ready-to-run program.md templates that optimize your cold email, landing pages, and pricing — on autopilot, at near-zero cost with local models.
12 questions, one honest score — sent straight to your inbox. No card.
Take the free assessmentAlready sold? Skip to the full playbook — $97 →
Autoresearch is powerful. But locked behind code.
The open-source repo needs a GPU, PyTorch, and Python. We translate the exact same optimization loop for marketing and sales operators — no terminal heroics required.
Built for ML, not ops
You can't point a GPU training loop at a landing-page headline or a cold-email subject line. This playbook bridges that gap with business-ready instructions.
Past the vague theory
Other guides explain what loops are. We hand you the exact templates: copy, fill in your metrics, and run on Claude Code or Cursor.
Complexity, in plain English
Every loop is framed in business terms: define what can change, pick the success scorecard, set the safety limits, keep the winner, repeat.
A structured operating handbook.
Everything you need to launch and run compounding AI loops — without writing a single script.
12 program.md templates
Ready-to-run instruction sets for local or cloud agents — covering cold email, headline testing, pricing realization, and onboarding activation.
Local-model cost guide
Route your loops to local models (Llama 3, Mistral, Gemma) via Ollama and MLX, and run unlimited experiments at zero marginal API cost.
7 failure modes
The exact ways agent loops waste money — changing two variables at once, undersized samples, optimizing vanity metrics — and how to avoid each.
Inspect a real program.md.
This is what an instruction set looks like. Switch tabs to see how the same loop adapts across domains.
2
3
4
5
6
7
8
9
10
11
12
13
14
Autopilot has a price tag. Local loops keep your cost at $0.
Run an optimization loop unattended on a paid API and it bills you per token — every iteration, every night. If you run loops nightly, here's the real math:
100 iterations a night at Opus 4.8 rates (~$5 / $25 per million tokens), metered per token, every night.
Run the identical loop on local models (Llama 3, Mistral, Gemma) on your own machine or Apple Silicon.
From zero to your first loop.
No setup marathon. Three moves and the agent is running.
Drop in a template
Pick the domain with the fastest feedback — usually cold email or landing pages — and paste the matching program.md into a file.
Add your numbers
Replace the bracketed placeholders with your current metrics, your offer, and your sending tool. Specific inputs, sharper output.
Walk away
Tell Claude Code or Cursor to read program.md and start the loop. Come back in 24–48 hours and feed it the results.
DAY
Run a loop. If it doesn't earn its keep, you don't pay.
Try the templates for two weeks. If you don't ship at least one optimization you'd have missed, reply to your receipt and we refund 100% — no questions, no conditions.
Get the Autoresearch Playbook.
One-time payment. Lifetime updates. Pick the bundle that matches how much hand-holding you want.
Starter set — the 3 core templates.
- 3 core templates — cold email, landing page, pricing
- Clean, ready-to-edit markdown
- Free lifetime file updates
- Setup guides not included
14-day money-back guarantee
Complete guide + all templates.
- All 12 ready-to-deploy program.md templates
- 28-page premium playbook (PDF)
- Local-model cost-savings appendix (Ollama / MLX)
- Model selection & routing matrix
- 7 documented failure modes to avoid
- 30-day execution roadmap
14-day money-back guarantee
Fix the prompts, then automate the testing.
- Everything in the Full Playbook
- 77-page AI Prompt Engineering Playbook
- 1h 5m audio guide + 5 ready-to-use tools
- The free assessment, included
Instant download · free lifetime updates
Common questions
What is the autoresearch loop?
Autoresearch is the keep-or-revert optimization loop Andrej Karpathy open-sourced: an AI agent proposes a change, measures it against a scorecard, keeps the change if it beats the baseline and reverts it if it doesn't, then repeats. This playbook adapts that exact loop for marketing — pointing an agent at your cold email, landing pages, and pricing so it compounds small wins automatically.
Do I need to know how to code?
No. The whole playbook is built for non-technical operators. You write instructions in plain English (Markdown), and an AI coding agent — Claude Code, Cursor, and the like — handles execution. If you can fill in a template and export a CSV, you have every skill you need.
Do I need an NVIDIA GPU?
No. The GPU requirement is for Karpathy's original ML application. The business loops here run through AI coding agents and the tools you already use — your email platform, page builder, and ad dashboard. No GPU, no PyTorch.
What does it actually cost to run an autoresearch loop?
Interactive use — where you run the loop and review each step yourself — stays inside Claude Pro or Cursor at about $20/month. Fully-autonomous overnight loops bill per token through the API (roughly $6–$15 a session at Opus rates). Or route everything to local models via Ollama / MLX and run at zero marginal cost. The playbook walks through all three.
How is this different from Karpathy's autoresearch GitHub repo?
Karpathy's autoresearch repo is an ML research tool that needs Python, PyTorch, and a GPU. This playbook extracts the underlying pattern — the keep-or-revert optimization loop — and turns it into structured instruction templates for business: email outreach, landing pages, ad creative, pricing, and onboarding. Different audience, different application, same proven pattern.