8h ago

Victor Taelin shares a prompting technique for Codex that directs the model to generate its own task instructions and correct errors without further user input

Bojan Tunguz quote-tweeted with plans to test the method.

1143.7K1101.4K142.6K

——0——

Original post

I discovered a new joy in life. Don't ask Codex to do stuff. Ask Codex to ask Codex to do stuff. Rejoice as you watch it handling and correcting all the dumb shit that it does and that you'd be dealing with otherwise

2:07 PM · May 21, 2026

QUOTE POST

#687Bojan Tunguz@TUNGUZ

I need to try this.

Taelin@VictorTaelin

9:07 PM · May 21, 2026 · 138.6K Views

11:24 PM · May 21, 2026 · 6.4K Views

Sentiment

Pos79.2%

Neg20.8%

Many users are excited by the meta prompting trick that lets Codex self-correct its own code because it feels like an elegant and satisfying way to outsource debugging, while others warn it risks wasting tokens and creating costlier errors.

24 comments with sentiment.

Digg Depth

Jonny Dasked

How well does this meta-prompting trick scale for complex, multi-file projects beyond simple Codex tasks?

The meta-prompting trick scales promisingly to moderate complexity by nesting sub-agents that plan, delegate, and self-correct across files, with users reporting big productivity lifts from manager-worker setups. Evidence stays anecdotal for large multi-file projects, where error compounding and coordination overhead could limit gains without extra scaffolding.

Cluster engagement

40 snapshots