If you read yesterday's piece on AEO/GEO and AI traffic in GA4, set it aside for a second — because this week the term "AEO" is being used by an entirely different audience to mean an entirely different thing. On April 11, Addy Osmani, Director of Engineering at Google Cloud AI, published a long-form piece coining Agentic Engine Optimization (AEO): the practice of structuring, formatting and serving technical content so that AI coding agents can actually use it — not human readers, not marketing answer engines, but Cursor, Claude Code, Codex, Aider, Amp, Windsurf, and the swarm of MCP-enabled tools sitting on top of them. Within a week the open-source agentic-seo CLI was on GitHub auditing sites for it. This is the same acronym as the marketing term, aimed at a completely different consumer.
1. Why this matters now
When a developer asks an AI coding agent "how do I integrate Stripe webhooks in Laravel?", the agent does not click through to your beautiful Astro-generated docs site, render your CSS, scroll the sidebar, and hover the copy-code button. It fetches a URL, tokenises the response, and decides in milliseconds whether your content is worth burning context window on. If your homepage returns a 500KB JavaScript-rendered shell with the actual content lazy-loaded by a hydration script, you are invisible. If your robots.txt blocks GPTBot, ClaudeBot, PerplexityBot or Google-Extended (often by accident, often inherited from a 2024 template), you are invisible. If your docs require login to view, you are invisible. The agent quietly recommends your competitor instead, and you never see the lost integration.
2. The five-signal AEO playbook
Osmani's framework — and the open-source audit tool — checks for five things. First, robots.txt: explicitly allow the agent user-agents you want. The default deny that many CDNs ship with is now an active business risk. Second, llms.txt: a markdown index at your domain root listing your most important pages, each annotated with a token count. This is the "table of contents" agents read first when deciding what to fetch.
Third, AGENTS.md: a per-repository file (now stewarded by the Linux Foundation's Agentic AI Foundation, with OpenAI, Google, Cursor, Amp and Aider all aligned) that tells coding agents how to build, test, and contribute to your project. Over 2,500 GitHub repositories already ship one, and GitHub's own Copilot guidance now treats it as a first-class signal. Fourth, Markdown twins: for every important HTML page, publish a .md mirror at a stable URL. Quick-start pages should fit under 15K tokens, API references under 25K, conceptual guides under 20K. This is the new "mobile site" — same content, agent-friendly format. Fifth, token efficiency: strip the marketing fluff, the cookie banners, the "you might also like" carousels. An agent that has to spend 8K tokens to find one code snippet will pick the competitor that gave them the snippet in 600.
3. What developers should ship this quarter
If you maintain an SDK, an API, or any developer-facing product, this is a one-sprint job with disproportionate payoff. Add an AGENTS.md to the repo. Add llms.txt to the docs domain. Add a /llms-full.txt containing the entire docs concatenated for one-shot ingestion. Re-check your robots.txt. Run agentic-seo audit and fix what it complains about. The teams that ship this in Q2 2026 will be the ones AI agents recommend by default for the next 18 months — and "default recommendation" is the new top-of-funnel.
My Take
There is a curious symmetry between this week's "Mini Shai-Hulud" supply chain attack and Agentic Engine Optimization. Both stories are about a world in which AI coding agents are now the primary reader, the primary integrator, and — increasingly — the primary attack surface of developer-facing software. If 2024 was the year we learned to prompt these agents, 2026 is the year we learn to publish for them. The repos that figure this out first will quietly become the new defaults; the rest will spend the next year wondering why their integration numbers fell off a cliff and "we didn't change anything."
Sources
- Agentic Engine Optimization (AEO) — Addy Osmani
- agentic-seo — addyosmani/agentic-seo on GitHub
- Agentic engine optimization: Google AI director outlines new content playbook — Search Engine Land
- AGENTS.md — official site
- How to write a great agents.md: Lessons from over 2,500 repositories — The GitHub Blog