According to an OpenAI announcement dated August 18, 2026, work-management company Asana used OpenAI Codex to remove Enzyme, an outdated testing system that OpenAI says had fallen out of active maintenance and was blocking Asana from modernizing its frontend stack. OpenAI states the migration was expected to take at least five years under the previous staffing plan, at an estimated cost of roughly $6 million, and was instead completed in about two weeks. OpenAI reports the model and infrastructure costs totaled about $12,000, with roughly 1.5 weeks of engineering effort spread across two calendar weeks. Per the account, an engineer wrote a five-sentence prompt, after which up to four coding agents worked in parallel, each in a separate copy of the codebase. An engineer checked progress twice a day and reviewed every proposed change. OpenAI notes that simpler instructions worked better than a more elaborate setup. OpenAI states the experience changed which long-running software projects Asana believes are practical to take on, and that Asana plans to test agents on other migrations, rewrites, and performance problems. This is a vendor-published case study; the figures are attributed to OpenAI and Asana and have not been independently verified.
- OpenAI reports the Enzyme testing-system removal took about two weeks versus an estimated five years.
- Stated model and infrastructure cost was about $12,000, against a roughly $6 million prior staffing estimate.
- Up to four Codex agents worked in parallel from a five-sentence prompt, with an engineer reviewing every change twice daily.
What it means for you
A large software company says AI coding agents did a tedious, well-defined cleanup job in weeks that engineers had budgeted years for. The headline numbers come from the vendor selling the tool, so treat them as a marketing claim, not a proven benchmark. Still, the underlying pattern is real: repetitive, large-scale code changes with clear rules are exactly what these agents handle well, and a human reviewed every change.
Try this
If you or your team maintains code, pick one boring, well-defined migration or cleanup task you've been putting off, and try scoping it as a short, plain-language prompt for a coding agent — then have a person review every change before merging.
Who should care
Engineering leaders and developers sitting on large 'someday' code-migration or cleanup projects they've assumed were too expensive to start.
Skip this if
You don't write or maintain software, or you don't have engineers who can review AI-generated code changes before they ship.
Sources: OpenAI — read the original