Hey PH 👋 Raymond from Synscribe. Six months ago we started asking a simple question: when an AI coding agent "decides" to use Stripe or Resend or Postgres, what actually made that choice? We expected to find SEO-with-extra-steps. We found a different machine entirely. Some of what we measured, building this: > A two-year-old auth library is the #2 most-fetched docs source among coding agents, ahead of React. Defaults are NOT locked into training data. > In our pilot experiments, a one-paragraph AGENTS.md file flipped an agent's product choice 100% of the time, and the agent invented reasons to praise the product it was told to use. > Bun's growth hack is a filename: `bun init` writes `use-bun-instead-of-node...` rules into every project it scaffolds. > We instrumented agents doing real research: they opened ~6% of search results and killed ~48% of vendor claims they fact-checked. > ~99% of MCP distribution bypasses the directories everyone tells you to list in. So we wrote the playbook we wished existed: 11 plays across the agent's whole journey, 10 company teardowns (Stripe, Bun, Tailwind, Better Auth…), a 30-min self-audit, and a mythbusting ledger where we publish "unproven" out loud when our data can't settle it. Everything is free, ungated, with methodology and limitations published. Our experiments are pilot-scale and labeled that way on every page. The meta part: the guide practices what it preaches. It ships its own llms.txt, serves every page as raw markdown, and stamps every claim with a verification date. We ran the playbook on the playbook. The data came from an agent-observability instrument we built (think: Search Console for the part of the funnel where agents decide). I'll be here all day. Bring skepticism.

The Agentic Discovery Playbook

GEO Playbook for Optimizing Products for AI Agents

GEO Playbook for Optimizing Products for AI Agents