Skip to content
How to ThinkIn the Age of AI

Question

What is agent interface design?

beginner

Quick Answer

Define how the output of one agent becomes the input of another.

Example: You have a morning routine with three agents: a planning agent that decides what to work on, an execution agent that does the work, and a review agent that evaluates the output. Each agent does its job well in isolation. But the planning agent produces a vague intention ('work on the project'), the execution agent interprets it differently than intended, and the review agent evaluates against criteria the planner never specified. Three capable agents, zero coordination — because nobody defined the protocol. The output of one agent does not cleanly become the input of the next. Compare this to a system where the planning agent outputs a structured brief (task, success criteria, time constraint, context from yesterday), the execution agent receives that brief and produces a deliverable tagged with which criteria it addressed, and the review agent scores the deliverable against the original criteria and passes a structured assessment back to the planner. Same three agents. But now they speak a shared protocol, and the system actually works.

This concept is part of Multi-Agent Coordination in the How to Think curriculum.

Learn more in these lessons