For as long as documentation has existed, it has had one assumed reader: a human being who is stuck and trying to get unstuck. Every decision we make as technical writers — the heading structure, the example we choose, the warning we place before a destructive command — is made with that human in mind.
That assumption is quietly breaking down.
Agentic AI systems are increasingly the first consumers of documentation. When a developer asks an AI coding assistant to set up authentication, the agent reads your getting-started guide, interprets your API reference, and acts on your error messages — often without a human ever seeing the page. When an AI-powered support tool answers a user question, it is synthesising your documentation and presenting a version of it that you did not write and cannot review. The human is still downstream, but they are no longer the reader.
This raises questions that the documentation community has not seriously grappled with yet. How does writing change when your reader has no patience for narrative context but can parse dense reference material perfectly? When your reader cannot tell you it is confused, cannot infer what you meant, and will act on what you literally wrote rather than what you intended? When optimising for agent readability and optimising for human readability pull in opposite directions — which reader do you write for?
This talk is speculative but grounded in what is already happening. Drawing on real examples of how agentic systems consume and misuse documentation today, we will examine what this shift means for how FOSS projects structure their docs, what it means for the humans who still need to read them, and whether the documentation community risks optimising itself into irrelevance by chasing a moving target.
The goal is not to alarm but to provoke honest thinking about a transition that is already underway — and to ask what documentation is actually for, before the answer gets made for us.
