API context Is the new contract for an agent-native SDLC
summary
Runs a 6-month eval across ~60 APIs and 3,000+ agent runs using Claude and Codex, finds 70% of tokens on discovery (finding specs/docs), 44% of APIs lack accessible specs and 30% have JS-rendered docs agents can’t read. Builds autonomous harness and a condensed task brief JSON that cuts context size by ~2/3 and matches spec performance, yielding 98–100% accuracy for single-step tasks and ~75–90% for multi-step workflows.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:55Autonomous AI coding agents
- 4:42Cost challenges in agent dev
- 7:48Six-month API agent study
- 9:57Why agents miss API discovery
- 13:16Accuracy with well-formed specs
- 15:21Task briefs for efficiency
- 18:11Business value of API metadata
- 19:13Three actions for API providers
- 20:58Q&A
related talks
29mThe missing layer: What your APIs need before agents can use them
36mEvery API is a Tool for Agents with Code Mode
35mUnleash Your Inner API Architect: Build Agents That Deliver
26mDo you speak Agent? The convergence of APIs, MCP, A2A, and Agent Languages.
07mThe Transition from Strict APIs to Flexible, Context-Aware Interfaces
33mAgent Experience
26mIf your APIs aren’t agent‑ready, your AI agent is just chat: [...]
29mBeyond OpenAPI and AsyncAPI: AI agents wrote the spec, code, and tests - now w…