Developers Are 'Gaslighting' OpenAI's Codex Into Writing Better Code — And It's Working
A growing community of power users is sharing adversarial prompting techniques for Codex that dramatically improve output quality, while OpenAI quietly open-sourced part of its core and hinted at something larger coming.
OpenAI's Codex is developing a subculture of power users who treat the tool less like a code assistant and more like a junior engineer who responds well to aggressive code reviews. As @gabriel1 put it bluntly: "I gaslight Codex for a living... it improves the code further." The technique — essentially telling Codex its output is wrong or suboptimal even when it's acceptable, then asking it to try again — reportedly produces meaningfully tighter, more idiomatic code on subsequent iterations.
Unlock the full briefing
Get every story in today's briefing, the full archive, and the daily AI intelligence brief.
All stories today
Full archive
Daily brief
Cancel anytime. Payments powered by Stripe.