Wire flash
ReadMe: 72% of traffic on its API docs now comes from AI agents
Editorial responsibility
- No named human review is recorded for this page.
- Source reporting is collected, normalized, translated or condensed automatically when needed.
- Automatically published source-backed update
An X post argues that AI assistants, not humans, are increasingly the first readers of API documentation and will decide within seconds whether to use a product or a competitor's. The post cites that 72% of traffic on ReadMe's API docs now comes from AI agents. It notes that when an agent hits an error, it reads only the error message; a vague "invalid request" response becomes a dead end and the agent simply picks another tool, producing no support ticket and unexplained flat signups. The post says error messages are now a critical product surface, citing Restless for embedding fix steps and doc links in failed responses. It argues APIs were built for long-lived integration contracts, while agents need fast-changing, agent-facing layers covering settings, sign-up, and permissions. Sam is quoted saying every company is an API company now, and the post notes agents' feedback can now land as Linear issues.
Source report
Their AI assistant will show up, read a page of instructions meant for programmers, and decide in a few seconds whether to use your product or a competitor's. Nobody on your team will know a decision was even made.
The Shift Is Already Here
This is already happening. 72% of the traffic on ReadMe's API docs now comes from AI agents. Those docs are the manuals software companies write for developers — and most of their readers stopped being developers.
What Happens When Something Breaks
Think about what happens when something breaks:
- A person who hits an error Googles it, asks a coworker, pokes around the dashboard.
- An agent reads the error message and nothing else.
If it says "invalid request," that's a dead end — and an agent at a dead end just picks another tool. No support ticket, no angry email. Your signups get a little flatter and you never find out why.
So the error message is now one of the most important screens your product has. Restless puts the fix steps and a doc link right inside the failed response, which sounds minor until you realize that's the one spot an agent is guaranteed to read.
The Deeper Problem: How APIs Were Built
The deeper problem is how APIs got built in the first place. They were made for integrations: a developer reads the docs once, writes the code, ships it, and that setup has to keep working for years. So companies only exposed what integrations needed. Settings, sign-up, and permissions all stayed behind buttons in the UI.
Agents need all of it. I hit this constantly with my own agents. They get 90% of a task done, then hand me a link to some settings page so I can click one toggle. The human ends up as the agent's courier.
Sam put it bluntly: every company is an API company now, whether they want to be or not.
Two Layers, Two Timeframes
An integration contract has to last for years. An agent contract only has to last a few seconds. Once you see that, the smart move is two layers:
- A frozen API for integrations
- An agent-facing layer that changes as fast as the product does
Closing the Feedback Loop
The last piece flips the feedback problem. Agents were always going to know exactly what's broken in your API — they just had nowhere to say it. Now a stuck agent gets asked what went wrong, and the answer lands as a Linear issue. Somebody finally built the inbox.
Source
aakashguptaNeutral / independent
Part of this Story
ReadMe Launches Restless as 72% of Its Documentation Traffic Comes From AI Agents