Working with AI agents
Give an agent durable product context without sending it on a scavenger hunt.
An AI agent needs the product’s intent, constraints, and interfaces—not an oversized prompt full of stale implementation detail.
Give the agent a stable entry point
Start with /llms.txt. It lists each public document with a short description. Use /llms-full.txt when a task needs the entire public documentation set in one Markdown file.
Inside the docs UI, Copy for AI adds page title and source frontmatter before copying the raw Markdown. Each section also has a copy control, so an agent can receive the smallest useful slice of context.
Product context for a coding task
Provide the agent with these files before asking it to change the product:
AGENTS.md
docs/adrs/0008-vercel-platform-first.md
apps/test3/content/site.yaml
apps/test3/content/docs/<relevant-page>.md
Then name the user outcome and the boundary. For example: “Add a public comparison page; do not add a background worker or a shared package.”
Keep generated apps legible
Use YAML for copy and Markdown for documentation. Keep decisions, schema, routes, and deployment code explicit. An agent should be able to trace an outcome from content to component to deployment without guessing which hidden generator rule changed it.