What's this about?
OpenAI published a case study on Asana using Codex to remove Enzyme, an outdated frontend testing framework that had made its codebase hard to upgrade, and migrate to React Testing Library. The work took about two calendar weeks against a plan that estimated five years. Model and infrastructure costs were roughly $12,000, against Asana's own estimate of roughly $6 million for the previous staffing plan. Asana's engineers reviewed and approved each proposed change. Their stated caveat is that not every years-long project will collapse into weeks.
What exactly is a test migration, and why does it take five years?
Enzyme and React Testing Library are both libraries for testing user-interface code, but they embody opposite philosophies — Enzyme reaches into a component's internals, while React Testing Library tests what a user would see and do. So migrating isn't a find-and-replace; every test has to be rewritten to check different things. A large application has tens of thousands of them. Any individual test takes maybe twenty minutes and is completely mechanical. Multiply that by the count and you get a project that is enormous, boring, generates no new features, and therefore never gets funded — which is exactly why it sat undone for years.
What's the community saying?
- The cost ratio is what makes this different from a testimonial. $12,000 against $6 million is a 500x difference, and it's an unusually specific number for a vendor case study.
- The task shape explains the result, and limits it. This is the ideal case for an agent: mechanical, enormous, individually verifiable, with tests as the oracle. It says little about work requiring judgment.
- Asana's own hedge deserves the credit it got — "not every years-long project will collapse into weeks" is the sentence a marketing team would normally cut.
- The five-year figure is a plan, not an observation. It's what the staffing plan projected, which is a real number for budgeting purposes but not the same as five years of work actually done.