Simon Willison, a prominent voice in the AI and LLM community, released a comprehensive guide on agentic engineering patterns on March 4, 2026 documenting emerging patterns for effectively working with AI coding agents. The guide synthesizes best practices for human-agent collaboration in software development, emphasizing strategic decision-making over implementation details.
Code Generation Fundamentals Shift Development Focus
The guide introduces two foundational principles that reframe how developers should approach AI-assisted coding. First, "code is now inexpensive" — acknowledging that agents can generate code quickly, shifting developer focus from typing to strategic architectural decisions. Second, "preserve domain expertise" — developers should retain knowledge work and focus agent activity on routine implementation tasks.
Test-Driven Development With AI Agents
Willison advocates applying red/green TDD methodology specifically adapted for agent workflows. The recommended approach starts with failing tests before prompting agents to generate code. Developers should run existing test suites before engaging agents to establish baseline behavior, ensuring agents understand current system state and expected outcomes.
Code Comprehension Through Interactive Dialogue
The guide recommends using linear walkthroughs to break down code comprehension into sequential steps. Rather than asking agents to explain entire codebases at once, developers should engage in dialogue to understand code behavior dynamically through targeted, interactive questions that build understanding incrementally.
Practical Examples and Prompt Collection
The guide includes annotated prompts demonstrating real-world applications, such as GIF optimization using WebAssembly. Willison provides a curated collection of effective prompts that serve as reference material for developers learning to work with coding agents like Claude Code and OpenAI Codex. These examples illustrate how to structure requests for maximum clarity and effectiveness.
Key Takeaways
- Simon Willison published an agentic engineering patterns guide on March 4, 2026, synthesizing best practices for AI-assisted software development
- Core principles include treating code as inexpensive while preserving human domain expertise for strategic decisions
- The guide recommends test-first development with agents, running existing tests before generating new code
- Linear walkthroughs and interactive dialogue enable better code comprehension than one-shot explanations
- The guide includes practical examples like GIF optimization and a curated prompt collection for reference