Five tools, Markdown, one LLM: making agents compound knowledge you own
dfinke · x · 2026-09-10
Doug Finke argues that most AI conversations are disposable — the agent does the work, then forgets everything it learned when the session ends. His experiment uses a tiny harness: five basic tools, Markdown files, and a single LLM, with no frameworks.
The core metaphor: your work becomes a library, and the AI just needs to find the right book at the right moment. The goal is to see how little machinery it takes before an AI behaves like it has a brain you actually own — continuously organizing, retrieving, and compounding knowledge you own while it works. The idea echoes Garry Tan's G-Brain ("the library plus the librarian") and Andrej Karpathy's LLM Wiki blueprint.
Related event: Doug Finke's Experiment: Giving AI a Compounding Personal Knowledge Base(3 posts)→
More from coding & agent
- Claude Projects now lets you drag threads into the message box or @ mention them for precise context — dfeinition · 2026-10-27
- Fern 0.1: a 0.1–0.5T coding model one researcher pretrained in 3.5 months for under $10M — teortaxesTex · 2026-10-27
- Jane Street's Yinsky shares tips for making codebases agent-friendly: types, property tests, formal methods — sh_reya · 2026-10-27
- Celesto launches CI runners nearly 90% cheaper than GitHub Actions, with agent sandboxes — aniketmaurya · 2026-10-27
- Runway launches App Studio: build no-code creative apps that generate image, video and audio — runwayml · 2026-10-27
- Supabase founders on rebuilding the backend for a billion AI-built apps — Latent Space · 2026-10-27