Simon WillisonSep 30, 2025

One of the new skills required to get the most out of AI-assisted coding tools - Claude Code, Codex CLI, etc - is designing agentic loops: carefully selecting tools to run in a loop to achieve a specified goal. Do this well and you can solve many coding problems with brute force

Here's my expanded explanation of what it means to design an agentic loop, how to do it safely (while running in YOLO mode!) and kinds of interesting problems this approach can be used to tackle https://simonwillison.net/2025/Sep/30/designing-agentic-loops/

Designing agentic loops

Coding agents like Anthropic’s Claude Code and OpenAI’s Codex CLI represent a genuine step change in how useful LLMs can be for producing working code. These agents can now directly …

Simon Willison’s Weblog
Show threadChris Zetter
@simon yes I found that feedback loops are key- I got Claude code to build a quite complex system by giving it lots of existing tests. Where I struggled was setting up good feedback loops on API design and architecture which caused the code quality to suffer (I wrote about it at https://technicaldeft.com/posts/can-coding-agents-build-complex-systems )
Can coding agents build complex systems?

What are the strengths and weakness of using AI coding agents for building complex systems

Sep 30, 2025 at 6:33pmWeb