FULL STORY

OpenAI Codex in Action: From Complaints to Rapid Refactoring

Despite initial complaints about its security tooling, OpenAI Codex impressed developers in real-world tests, demonstrating the ability to rapidly refactor complex codebases.

2026-07-30 ~ 2026-08-08 · 3 episodes · 9 posts

Episode 1 · OpenAI Codex Security Slammed for Poor User Experience (2026-07-30, 2 posts)

A developer complained on Reddit that OpenAI's Codex Security tool ran for 20 minutes, exhausted the weekly quota, and then refused to display the scan results. This poor experience has sparked concerns about the tool's overall practicality.

Episode 2 · Developers Test OpenAI Codex: Runs Old Game, Builds Panel in 5 Min, Cuts Code to 1/10 (2026-07-30, 5 posts)

Multiple developers recently tested OpenAI Codex and gave high praise. Riley Ralmuto called the experience 'like a polite and humble ASI'; gabriel1 used it to run a 25-year-old Windows 95 game; billyjhowell built a 3D printer monitoring panel in 5 minutes; joshbickett got code reduced to 1/10th size. Lucasmeijer thinks Claude Code will struggle to compete. These tests showcase Codex's strong capabilities in complex task handling, code generation, and optimization, sparking discussion on the competitive landscape of AI coding assistants.

Confirmed

  • Running old game: gabriel1 used Codex's Computer Use feature to successfully run a 25-year-old Swedish Windows 95 game. Initially the disc was recognized as an audio file, but the Codex agent solved the problem on its own.
  • Rapid app building: billyjhowell gave Codex an 'impossible' task before bed, and it was done before he shut down; then he asked for a 3D printer monitoring panel, completed in 5 minutes.
  • Code reduction: joshbickett asked Codex to halve the code size; Codex delivered a working version at 1/10th the original size.
  • Developer opinions: Riley Ralmuto called it 'like a polite and humble ASI'; lucasmeijer thinks Claude Code will have little competitive edge except for user habit.

Unconfirmed

  • All experiences are personal shares without reproducible steps or benchmarks; actual performance may vary by task.
  • The 'ASI' description is subjective, not an objective definition of Codex's capabilities.

Why it matters

These tests indicate OpenAI Codex excels at automated programming tasks, potentially reshaping the competitive landscape of AI coding assistants. If its capabilities match developer claims, it could challenge existing tools like Claude Code and push AI coding tools toward more autonomous and efficient development.

Episode 3 · OpenAI Codex Refactors Complex Codebase in Two Hours (2026-08-08, 2 posts)

A veteran developer was astounded after OpenAI Codex successfully refactored a highly complex, multi-year codebase from scratch in just two hours, demonstrating the AI's breakthrough capabilities in programming.