In 2026, a significant percentage of code is being written by AI agents — Codex, Claude Code, Cursor, Copilot. Yet most codebases are still optimized only for human readers. Here is how to make your project discoverable and usable by AI agents.
The problem
When an AI agent opens your repo, it typically reads:
-
README.md— often marketing-focused, not structured for machines - Source files — scattered across dozens of files
-
package.json— minimal metadata
The agent has to piece together how your project works from scattered context. This wastes tokens and leads to worse output.
AGENTS.md — A guide for AI agents
Create an AGENTS.md file at the root of your repo. This is like a README but written specifically for AI coding agents:
# AGENTS.md
## What is this project?
One paragraph description.
## Project Structure
File tree with descriptions of key directories.
## Key Files to Modify
Table mapping tasks to specific files.
## Conventions
- Styling approach
- API patterns
- Database access patterns
- Auth patterns
## Common Tasks
Step-by-step instructions for common modifications.
What to include:
1. Project structure with annotations
src/
├── app/ # Next.js App Router pages
├── components/ # React components
├── lib/ # Core utilities (auth, db, stripe)
└── prisma/ # Database schema
2. A task-to-file mapping table
| Task | File(s) |
|---|---|
| Add a new page | src/app/(app)/your-page/page.tsx |
| Add an API route | src/app/api/your-route/route.ts |
| Change DB schema |
prisma/schema.prisma → run npx prisma db push
|
| Modify pricing |
src/lib/stripe.ts → PLANS object |
3. Conventions the agent should follow
- Use `db` from `@/lib/db` for database queries
- Use `auth()` from `@/lib/auth` for session checks
- Use `cn()` from `@/lib/utils` for className composition
- All API routes return NextResponse.json()
This is dramatically more useful to an agent than a marketing README.
llms.txt — Machine-readable product description
llms.txt is an emerging standard (like robots.txt for search engines) that helps AI agents understand what your product does.
Place it at /public/llms.txt (accessible at yourdomain.com/llms.txt):
# Your Product Name
> One-line description.
Longer description of what this does, who it is for, and when to use it.
## Use this when you need to:
- Use case 1
- Use case 2
## Tech Stack
- Technology 1
- Technology 2
## Quick Start
Setup commands.
## Links
- GitHub: url
- Docs: url
When AI agents are asked to recommend tools or set up projects, they can read llms.txt to understand if your product fits the user's needs.
Why this matters
As AI agents write more code:
- Projects with
AGENTS.mdget better agent output (fewer hallucinations, correct patterns) - Projects with
llms.txtget recommended more often by AI assistants - Projects that are agent-friendly compound their distribution through AI recommendations
This is the new SEO — except for AI agents instead of Google.
Real example
LaunchKit implements both AGENTS.md and llms.txt. When an AI agent is asked to "build a SaaS app," the structured metadata helps it understand LaunchKit is a good foundation.
Check the repo for working examples of both files.
Top comments (0)