Скачать или смотреть How Speakeasy approaches API design for AI agents

In this interview, Georges Haidar, a software engineer at Speakeasy, explains how traditional REST APIs need to evolve for LLM agents. Learn about the key differences between designing APIs for developers versus AI agents, common pitfalls like runaway tool calling, and how Speakeasy's new product Gram helps bridge this gap. Ben and George also cover practical solutions including better API discoverability, context-aware design, and the role of MCP (Model Context Protocol) servers in the agent-first future.

Check out more at https://www.speakeasy.com/
Watch Georges demo Gram    • Building MCP servers using Gram from Speak...  
Watch Georges's Stripe developer meetup talk in London    • Bridging the gap between your API and LLMs  

Chapter Markers
00:00 - How API design changes for agents
01:44 - Speakeasy's approach to the problem
02:29 - Common LLM failures with unprepared APIs
04:13 - What Speakeasy does and the Gram product
05:25 - Understanding tools vs MCP servers
07:10 - Making APIs more discoverable for LLMs
09:11 - Retrofitting existing APIs for agents
12:26 - Rate limits and error handling for agents
13:52 - Chaining tools effectively
15:17 - The future of agent-driven API consumption