This talk was presented at the AI The Docs 2025 online conference. We are thrilled to share the recording and the summary with you.
Visit the talk summary page to see all of the presentations from the conference.
Summary
How will APIs and their documentation evolve in an AI-first world where agents integrate, use, and even create APIs?
Traditionally, APIs required human involvement for design, integration, and maintenance (“API plumbing”). In his keynote presentation, Zdenek Nemec (Founder and CTO of Superface.ai) underlines, we’re moving toward A2A (agent to agent), where AI agents connect directly. As he sees, AI is now able to:
- Integrate APIs
- Use APIs
- Build/deliver APIs, and
- Will conduct business on top of APIs
Demo highlight: Using Claude + Superface, an AI agent integrated with Todoist without explicit API descriptions: finding human-facing docs, writing integration code, testing against the live API, and deploying it. This raised a provocative question: do we still need formats like OpenAPI if AI can parse human-centric docs?
The reliability GAP: Despite the demo, AI agents succeed only ~25-55% of the time in repeatable CRM tasks.
Documentation is critical to close this gap by:
- Training LLMs on public API docs
- Fine-tuning with good/bad examples
- Providing structured tools and Model Context Protocol (MCP) servers
- Designing APIs with AI consumption in mind
Documentation challenges today:
- Business processes are rarely documented.
- Docs are often “out-of-band,” leading to version mismatches.
MCP and self-describing APIs: MCP provides a uniform way for APIs to share capabilities with AI. Supported by major LLM providers and platforms, MCP makes APIs self-describing: feeding updated documentation to agents at runtime, solving versioning problems.
The future of APIs:
- Now: human-centric APIs (AI scraping human docs).
- Next: self-describing APIs (enabled by MCP).
- Later: machine-only APIs, where human-driven integration is too slow.
Business implications: AI will increasingly perform B2B transactions via APIs, requiring documentation not just of endpoints, but of business processes, terms, and contracts.
Key takeaway: API documentation is no longer just for humans: it is training data, runtime context, and a prompt for AI agents. In the agentic future, documentation will directly determine the success or failure of AI-driven integrations.
Sign up to our Developer Portal Newsletter so that you never miss out on the latest API The Docs recaps and our devportal, API documentation and Developer Experience research publications.
