# How Alexander Embiricos actually uses coding agents

Codex product lead, OpenAI

**If you take one thing:** `"Using a subagent in parallel, do X"`, from one always-on thread. instead of opening new threads, he keeps one thread alive and hands every new task to a subagent from it.

- **Agent:** Codex (app) ([source, 2026-04-23](https://x.com/embirico/status/2047461700895662333))
- **Model:** Codex default model ([source, 2026-02-09](https://www.youtube.com/watch?v=6OiE0jIY93c&t=100))
- **Review:** Reads the plan; Codex reviews nearly every PR ([source, 2026-01-12](https://www.youtube.com/watch?v=xeZDHGjG5zM&t=1902))
- **At once:** 4 attempts ([source, 2026-01-12](https://www.youtube.com/watch?v=xeZDHGjG5zM&t=1375))
- **Harness:** One long-lived Codex thread, kept busy by automations, spawning subagents ([source, 2026-04-20](https://x.com/embirico/status/2046033614106931414))
- **Skills:** buildkite ([source, 2026-02-09](https://www.youtube.com/watch?v=6OiE0jIY93c&t=57))

## In their words

> Love using auto-review. It’s my new default mode

> especially because I'm often using [Codex] on low, which I think is like an under underrated technique.

> So I might kind of throw throw codex at it but I might just do it four times in parallel instead.

> Codex code review itself is enabled on pretty much every single repo at the company and reviews like pretty much all PRs.

> I'm slowly shifting to longer-lived threads. The thread is pretty much always active because I prompted it or an automation pinged it.


Archetype (our label): Orchestrator. Newest source: 2026-04-23.
Card: https://www.prehistoriq.com/stacks/embirico/ · Data: https://github.com/prehistoriq/stacks/blob/main/data/stacks/embirico.json