[How Coding Agents Work][02] Native Tool Calling and Structured Output

How Coding Agents Work · Module 02 of 29: Native Tool Calling and Structured Output Part 0 — Foundations How a model is given hands. Tool schemas as the model's only documentation, the shift from parsing free text to native structured calls, and the request-shaping choices that make a fifty-step run affordable. After this module you can write a tool schema a model will select correctly, and explain how request shape determines cache hit rate. Lessons in this module: 1. From text parsing to native tool calls: what changed, and what it eliminated 2. Tool schemas as contracts: names, descriptions, typed parameters 3. Structured output and strict JSON: results your code can consume 4. Responses API vs Chat Completions: statefulness, reasoning items, response chaining — and why Codex implements only the former 5. Prompt caching: cache keys, stable prefixes, and the economics of a long turn Prerequisites: module 1. About the course: Built from the OpenAI Codex harness at commit 8d01cd42fa.