# How Wes McKinney actually uses coding agents

pandas, Apache Arrow

**If you take one thing:** `roborev`, reviews every agent commit. he built a background reviewer that a second model runs on every commit his agents make, and will not code without it.

- **Agent:** Claude Code (cli) ([source, 2026-08-12](https://wesmckinney.com/blog/agentic-engineering-aug-2026/))
- **Agent:** Codex (app) ([source, 2026-05-08](https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html))
- **Model:** Claude Fable ([source, 2026-08-12](https://wesmckinney.com/blog/agentic-engineering-aug-2026/))
- **Model:** GPT-5.6 Sol ([source, 2026-08-12](https://wesmckinney.com/blog/agentic-engineering-aug-2026/))
- **Review:** roborev reviews every commit; he checks structure ([source, 2026-05-08](https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html))
- **At once:** 4-5 projects ([source, 2026-05-08](https://wesmckinney.com/transcripts/2026-05-08-show-us-your-agent-skills.html))
- **Harness:** Own stack: Ghosthub terminal over tmux/Herdr/Zellij, Kenn Forge worktree workspaces, Kata issues, roborev, AgentsView ([source, 2026-08-12](https://wesmckinney.com/blog/agentic-engineering-aug-2026/))
- **Rules file:** Clanker Constitution, linked into ~/.codex/AGENTS.md and ~/.claude/rules ([source, 2026-08-11](https://github.com/kenn-io/constitution/blob/48df1551c31fbbb783560e11006b61a43ef897fe/README.md#L12-L14))
- **Skills:** roborev-design-review, roborev-design-review-branch, roborev-fix, roborev-lookahead-review, roborev-lookahead-review-branch, roborev-refine, roborev-respond, roborev-review, roborev-review-branch, roborev-snooze, roborev-superpowers ([source, 2026-10-06](https://github.com/kenn-io/roborev/tree/4aa94982c61b31f24707d3740e9c7d62326bb42e/internal/skills/claude))
- **Hooks:** roborev post-commit review hook ([source, 2026-04-08](https://wesmckinney.com/transcripts/2026-04-08-joe-reis-ai-agents-mythical-agent-month.html))

## In their words

> Superpowers turns the reviewed spec into an implementation plan, which we implement either with subagent-driven development (mostly with Claude) or inline execution (mostly with Codex, because Codex’s subagents are… not great).

> I use both Codex and Claude to do development.

> The work produced by the latest frontier models (5.6-Sol and Fable) is extremely sloppy and almost never suitable for production without substantial hardening.

> The work produced by the latest frontier models (5.6-Sol and Fable) is extremely sloppy and almost never suitable for production without substantial hardening.

> I work on four or five projects during the day, I’ll run parallel spec interviews with superpowers.


Archetype (our label): Toolsmith. Newest source: 2026-10-06.
Card: https://www.prehistoriq.com/stacks/wesm/ · Data: https://github.com/prehistoriq/stacks/blob/main/data/stacks/wesm.json