Others LLM Archived

Why are library docs still written for humans when the first reader is now an agent?

Cursor / Claude Code / Devin are now the primary reader of any new SDK; doc sites front-load tutorials, marketing, navigation chrome and screenshots that waste agent context window and degrade tool-call accuracy.

Who has it: AI-native developers using agentic coding (Cursor, Claude Code, Devin, Copilot) on libraries and APIs that haven't shipped agent-grade docs.

8.3Score / 10

Severity8/10
Feasible today9/10
Market signal8/10
Competition gap8/10

Heard in public

“Why are people still telling me what to do? I don't want to do anything. What is the thing I should copy-paste to my agent?”

Quoted word for word from the public post linked below. Nobody submitted it to Real Problem AI.

Andrej Karpathy, Sequoia Ascent 2026 summary

What nothing does yet

Stripe Atlas for agent-grade docs: ingest a repo + OpenAPI spec, output a versioned, tested, agent-shaped doc bundle the maintainer can host. Per-library SaaS to maintainers ($99-499/mo).

Tools that solve this

4 named so far, and where each one stops.

Mintlify / ReadMe / GitBook
Beautiful human docs; AI-ready is a chatbot bolt-on, not structured agent contract.
Context7
Aggregates llms.txt at consumer side; doesn't fix source library docs.
llms.txt standard
Convention with no opinionated tooling, validation, or coverage SLA.
Anthropic docs MCP / OpenAI cookbook
First-party only.

Why it is archived

Retired in the 2026-10-25 freshness audit: llms.txt files and doc MCP servers (Context7 and others) now serve agents; the gap narrowed.

How you might build it

01OpenAPI + JSDoc / TSDoc parser
02Tested example generation
03llms.txt + MCP bundle builder
04Per-version diff + regression checks

Founders working in this area

Han Wang, Co-founder, Mintlify

Not affiliated with Real Problem AI.

#T14 · Canonical URL: https://www.realproblem.ai/archive/why-are-library-docs-still-written-for-humans-when-the-first-reader-is-now-an-agent