When most docs readers are AI agents, page size and accuracy matter more than design

When most readers of a docs site are software agents, small pages and correct facts count for more than good looks. Hahnbee Lee, who helped found Mintlify and serves as its technical chief, argued this on the AI Engineer show, in a chat with Lincoln Murr of Coinbase. An agent is an AI program that acts for a person.

What was said

Lee said Mintlify began in 2022, before ChatGPT, when docs work leaned on design. He still loves design. But he said it makes him a little sad that "the agents do not see your website and they do not care what it looks like or what the branding of it".

What counts now, he said, is page size, because a big page can fill an agent's context window, the limited text it can take in at once. The other need is accurate content. If an agent meets a minor snag during setup, it may try brute force, but "it can also just as easily churn to another product".

He also gave a number. Early this year, about 10% of viewership across Mintlify sites was agents. He said it is now closer to 80%. Human viewership is not falling, he said; agent viewership is rising much higher.

Why it matters

Our reading: a docs site is becoming an interface, and the test is whether an agent can finish setup. Looks and human visit counts say little about that.

The mechanism is simple. A bloated page takes up part of the agent's limited text budget. A stale fact causes a snag, and no human may be there to fix it. The agent may simply move on to another product.

Lee called automations that stop docs going stale very important. He also said software companies should keep investing in APIs, the plain interfaces machines call directly.

The other side

The 80% figure is Mintlify's own tracking across its sites. The excerpt does not say how agents are told apart from people. It is one vendor's view, not the whole web.

The churn point is Lee's description of how agents can behave. He gave no data on how often agents switch after a snag.

Lee was open about what he had not settled. He said he is still figuring out AI writing, that creative prose needs a human touch, and that judging whether docs are good is more subjective.