# How Thibault Sottiaux actually uses coding agents

Codex & ChatGPT at OpenAI

**If you take one thing:** `Random-file bug hunt`, automation, several times a day. a scheduled Codex automation literally runs Python random to pick a file, then finds and fixes a latent bug in it.

- **Agent:** Codex (app) ([source, 2026-06-20](https://x.com/thsottiaux/status/2068144722460475527))
- **Model:** GPT-5.6 Sol ([source, 2026-08-07](https://x.com/thsottiaux/status/2085597685948813610))
- **Model:** GPT-6 Astra ([source, 2026-09-08](https://x.com/thsottiaux/status/2097193293532848288))
- **Review:** Reviews what the agent presents; uses a fast model to understand code, never delegates understanding ([source, 2026-02-18](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent))
- **At once:** 4 tasks ([source, 2026-02-18](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent))
- **Harness:** Codex app with scheduled automations driven by skills; one fast in-flow thread ([source, 2026-02-18](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent))
- **Skills:** ask-questions-if-underspecified, daily marketing-research skill ([source, 2026-01-01](https://x.com/thsottiaux/status/2006624682515247604))
- **MCP servers:** linear ([source, 2026-02-18](https://every.to/podcast/transcript-how-openai-s-codex-team-uses-their-coding-agent))

## In their words

> These days I spend more time in the codex app than all other apps combined on my Mac.

> I talk to it for 5 minutes straight with things that just seem to require weeks of work, get up to get something in the fridge, pet the dog, come back and ... it's done.

> Looking at this 28 page deck right now where Codex is keeping track of all the individual launches. Each one of them I can't wait to see what people do with it.

Brought to you by an incredible team (and Astra)

> I’d been optimizing for long-running, multitasking workflows—this task will take 10 or 15 minutes, I’ll kick off four different things and come back.

> You cannot delegate understanding—you’re trying to get to understanding something, and speed there is a real advantage.


Archetype (our label): Orchestrator. Newest source: 2026-09-08.
Card: https://www.prehistoriq.com/stacks/tibo-openai/ · Data: https://github.com/prehistoriq/stacks/blob/main/data/stacks/tibo-openai.json