# How Simon Willison actually uses coding agents

Datasette, Django co-creator

**If you take one thing:** `Showboat`, agents demo their own work. a CLI he built so agents record the commands they ran and the real output, as proof the code works.

- **Agent:** Codex (app) ([source, 2026-09-04](https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L99))
- **Agent:** Claude Code (web) ([source, 2026-09-04](https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L101))
- **Model:** GPT-6 Sol ([source, 2026-09-22](https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/))
- **Model:** Claude Opus 5.5 ([source, 2026-09-22](https://simonwillison.net/2026/Sep/22/opus-and-sol-and-luna/))
- **Review:** Cross-model review: Claude checks GPT, GPT checks Claude ([source, 2026-07-05](https://simonwillison.net/2026/Jul/5/sqlite-utils-fable/))
- **Harness:** Codex desktop + Codex remote on laptop and DGX Spark, driven from his phone ([source, 2026-09-04](https://github.com/simonw/monthly-newsletter-archive/blob/1cc46b5c2012ad6319c4a40e3f73b049f199c5c0/2026-08-august.md?plain=1#L103))
- **Rules file:** Short per-repo AGENTS.md (CLAUDE.md just imports it) ([source, 2026-05-30](https://github.com/simonw/research/blob/ba0cb97b42da153c13c465fb617f5c45f1f03b04/AGENTS.md#L3))
- **Skills:** cogapp-markdown, setup-to-pyproject, upgrade-actions, uv-tdd ([source, 2026-02-24](https://github.com/simonw/skills/tree/a6579982ed89adabb9b72ddaad0cf854e8e2218b))

## In their words

> I continue to spend most of my time in the Codex macOS desktop app, now renamed (confusingly) to ChatGPT.

> I continue to prefer Claude Code for Web - now using Fable 5.1 - for asynchronous cloud tasks.

> I’m now using GPT-6 Sol and Claude Opus 5.5 as my default models in Codex and Claude Code.

> I’m now using GPT-6 Sol and Claude Opus 5.5 as my default models in Codex and Claude Code.

> The problem is it really does work—I’ve started habitually having Anthropic’s best model review OpenAI’s work and vice versa, because I’ve had that turn up interesting results often enough to be valuable.


Archetype (our label): Toolsmith. Newest source: 2026-09-22.
Card: https://www.prehistoriq.com/stacks/simonw/ · Data: https://github.com/prehistoriq/stacks/blob/main/data/stacks/simonw.json