Asana used OpenAI Codex to finish a five-year, six-million-dollar engineering migration in two weeks for about twelve thousand dollars, and that changes which projects are worth starting.
Asana just told OpenAI, on the record, that a testing-system migration it planned to staff for five years and roughly six million dollars got finished in two weeks for about twelve thousand dollars. That's not a benchmark score buried in a research paper. That's a line item that used to require a multi-year headcount plan and now requires an invoice you could put on a corporate card.
Here's the setup. Asana's engineering org runs on the same agent-heavy approach the company sells to its customers: humans set direction, AI agents do the grinding, humans review before anything ships. The team had one of those years-long albatross projects sitting on the roadmap, ripping out Enzyme, an old JavaScript testing framework that had fallen out of maintenance and was blocking the company from modernizing its frontend stack. The kind of project every engineering org has. Everyone agrees it needs to happen, nobody wants to staff it, so it sits there for five years accruing interest as technical debt.
Instead, Asana handed the job to OpenAI Codex. From a five-sentence prompt, up to four coding agents worked in parallel, each in its own copy of the codebase. An engineer checked in twice a day and reviewed every proposed change before it merged. Total engineering effort came out to about a week and a half, spread across two calendar weeks. Model and infrastructure costs landed around $12,000, against a staffing estimate of roughly $6 million for the original five-year plan. Asana's CTO, Amritansh Raghav, put it plainly: "Not every years-long project will collapse into weeks. But agents can give engineers more room for craft, and make once-impossible work worth attempting."
That last clause is the part a CMO or a founder should actually sit with. It's not "engineers got faster." It's "the list of projects worth starting just got longer." Every company has a shelf of work that got shelved because the return on five years and six million dollars of staffing never penciled out against the roadmap. If that same class of project now costs five figures and a couple of weeks, the shelf empties out fast. Asana's own framing backs this up, the company says that with Enzyme gone, it's now testing agents against other migrations and rewrites it previously assumed were permanent fixtures.
Asana isn't the only data point OpenAI is pointing to here. On Codex's product page, Sierra's engineering team says they now "ship in a weekend what previously took a quarter," and Harvey reports Codex cut its early iteration time by 30 to 50 percent. Different companies, different projects, same shape: work that used to require a quarter of calendar time and a staffed team now gets planned in days.
Here's the honest caveat, because this is exactly the kind of result that invites lazy extrapolation. Enzyme removal is a well-bounded technical problem, there's an existing test suite to prove nothing broke, a clear definition of done, and no ambiguity about what "finished" looks like. That's the profile of work agents are good at right now: mechanical, high volume, verifiable against a spec that already exists. It is not the profile of "figure out what our product should do next" or "resolve a disagreement between two VPs about architecture." The six million dollar figure is also Asana's own internal staffing estimate for a project that never actually ran, not an audited historical cost, so treat it as directional rather than as a receipt. And this still took a human checking in twice a day and approving every change. Nobody walked away for two weeks and came back to a finished migration.
None of that undercuts the number. It just means the honest translation isn't "AI replaces engineering," it's "the cost of attempting the migration you've been avoiding just dropped by two orders of magnitude." That's a different, more useful sentence, and it's the one worth carrying into your next roadmap meeting.
The scariest part of this story isn't the twelve thousand dollars. It's that Asana now has to go find the next five-year project sitting on its shelf, because it just proved the shelf is cheaper to clear than to keep.