APIs in Agentic world
summary
Claims that agent traffic to sites rose ~1,200% year‑over‑year (about 200% month‑over‑month) and that commoditized LLMs make model choice non‑differentiating lead to a focus on APIs as the tool layer. MCP standardizes agent‑to‑tool calls so a single MCP server can expose many endpoints, but real systems hit problems: Twilio‑scale suites (~1,400 endpoints) confuse LLMs, token limits and throttling break chains, and auth chains complicate flows. Practical fixes: embed LLM context, include examples, error patterns, versioning, response‑size limits and custom traits; Amazon Bedrock is cited as an LLM/agent host.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 2:14AI agents impact on APIs
- 6:31AI agent reliance on APIs
- 8:51Model Context Protocol MCP
- 12:02LLM challenges with APIs
- 16:32Mobile era API lessons
- 19:04Making APIs LLM-friendly with documentation
- 22:22API readiness recommendations
- 24:59API readiness for agents
related talks
27mAre APIs becoming obsolete? Think MCP, ACP, A2A, AGNTCY etc.
21mAre APIs becoming obsolete? Think MCP, ACP, A2A, AGNTCY etc.
26mDo you speak Agent? The convergence of APIs, MCP, A2A, and Agent Languages.
45mFrom Tools to Partners: The Rise of Agentic AI
24mHow Generative AI Apps Use APIs
36mPutting AI Agents to Work: A New Era of Open Source Connectivity with MCP
24mAgentic API - the only API you will ever need
28mAI and the state of APIs in 2026