red/green TDD

Overview

Red/green TDD is a test-driven development pattern adapted for coding agents. In this workflow, an agent is prompted to write tests first, run them to verify they fail (the “red” phase), implement the code needed to satisfy those tests, and then rerun the tests to verify success (the “green” phase). In the agentic coding context, this turns a broad coding request into a tighter execution loop with explicit checkpoints.

For AI Product Managers, this matters because it is a practical reliability pattern for autonomous or semi-autonomous software generation. Rather than asking an agent to produce a large feature in one shot, red/green TDD encourages incremental validation, clearer acceptance criteria, and easier debugging. It has emerged as part of broader agentic engineering practices discussed by builders using tools like Claude Code and GPT-5.4, especially as coding agents became more dependable at following instructions.

Key Developments

• 2026-02-23 — Simon Willison’s Agentic Engineering Patterns was highlighted as a guide for building and operating agentic systems, with red/green TDD cited as one of the specific coding-agent patterns it collects.
• 2026-04-03 — A newsletter mention described the operational prompt for red/green TDD: instruct agents to write tests first, run them to confirm failure, implement code, and rerun tests to confirm success. This framed the pattern as a concrete way to improve productivity and reliability with coding agents such as Claude Code and GPT-5.4.
• 2026-04-04 — Lenny Rachitsky shared Simon Willison’s view that late 2025 marked an inflection point for AI coding, with autonomous coding agents, benchmark progress, and thin templates helping make practices like red/green TDD more usable in real workflows.

Relevance to AI PMs

• Define clearer acceptance criteria for agent workflows. PMs can translate product requirements into testable behaviors, making it easier for coding agents to work against explicit pass/fail conditions instead of vague feature requests.
• Reduce risk in agent-generated code. Red/green TDD creates a built-in validation loop that helps catch regressions, hallucinated implementations, and misunderstood requirements before code is merged or shipped.
• Design more reliable human-in-the-loop processes. PMs evaluating AI developer tools can use this pattern to structure agent tasks into auditable steps, improving trust, reviewability, and team adoption.

Related

• Simon Willison — A key advocate of agentic engineering patterns, including red/green TDD for coding agents.
• Lenny Rachitsky — Amplified Willison’s framing of the AI coding inflection point and the importance of patterns like this.
• Claude Code — An example of a coding agent environment where red/green TDD can be applied.
• GPT-5.4 — Referenced as a model enabling more reliable coding-agent behavior, making iterative TDD loops more effective.
• Agentic Engineering Patterns — The broader pattern library in which red/green TDD is positioned.
• Test-driven development — The traditional software practice from which this agent-adapted pattern derives.