AI agents generated 66% of web traffic across documentation hosted on Mintlify in July 2026. By August the count reached 257 million agent requests against 131 million human page loads. The people who maintain that documentation now write for machines first, and they have a new job title ;knowledge engineer. Mintlify co-founder Hahnbee Lee posted the analysis on X this month.

Mintlify raised a $45 million Series B at a $500 million valuation in April 2026, led by Andreessen Horowitz and Salesforce Ventures, bringing total funding to $67 million. The company began with a $2.8 million seed led by Bain Capital Ventures in 2022, when the pitch was auto-generating docs from code. The pitch today is a knowledge layer for AI agents. Decagon, the customer support agent company whose product education lead appears in Mintlify’s report, raised $250 million at a $4.5 billion valuation in January, tripling its mark in under seven months. Sacra estimates Decagon reached $100 million in annualized revenue in July 2026. Capital is flowing to both ends of the pipe, where agents read and where agents answer yet knowledge quality sits in between, and at most companies nobody owns it.

On February 14, 2024, British Columbia’s Civil Resolution Tribunal found Air Canada liable after its chatbot told a customer they could claim a bereavement fare after completing travel. The airline argued the chatbot was a separate legal entity responsible for its own statements. The tribunal called that a remarkable submission and awarded $650.88 in damages. The finding that a company answers for every word on its website, including machine-generated ones, carries much further now that agents reproduce one bad answer across thousands of downstream interactions.

Mintlify’s report defines the knowledge engineer as the person who decides which source is authoritative, wires product releases and support tickets to content updates, publishes in formats agents can parse, including Markdown, llms.txt indexes and MCP tools, and measures where retrieval fails. Ethan Palm, who holds the title at Mintlify, describes the work as connecting information scattered across teams and turning it into something both users and staff can trust. David Hou, Product Education Lead at Decagon, frames it as a product requirement: agents need knowledge shaped and retrievable enough to finish tasks end to end, and the whole organization has to buy in or a source of truth never forms.

The report’s own numbers show how far the loop has automated. Between February and August 2026, users directed Mintlify’s agent to update documentation roughly 367,000 times , 95% of that volume ran through webhooks and cron jobs, and 61% of the resulting pull requests merged. In the accompanying survey, 83% of respondents say agents now draft their documentation updates. Only 9% let agents publish without review. Humans have moved from writing the pages to approving them.

HubSpot’s migration illustrates the failure mode the role exists to fix. The company found that AI models answering developer questions were pulling from legacy V1 API docs, current V3 docs and serverless examples at once, producing conflicting advice. After consolidating on Mintlify, HubSpot cut engineering resources spent on docs infrastructure by 50%, reallocating the time to developer growth. Coinbase launched an AI-native documentation site in six weeks and logged 8,704 AI assistant queries from 2,736 developers in the first 60 days, with MCP tool calls tracked alongside. Fabrizio Ferri Benedetti, Content Engineer at Elastic, described the split on June 12, 2026 as feeding the machines first, then guiding the humans, with writers curating context for both audiences.

For investors, the knowledge layer is now a category with a $500 million anchor and a $4.5 billion customer. Expect the next rounds to go to companies that measure where agents fail to retrieve an answer, and to tooling that sits between a codebase and every agent reading it. For founders, the Air Canada ruling sets the floor: if your agent says it, you said it. Knowledge engineering is the job of making that sentence safe.