A Five-Year Project Done in Two Weeks: Asana's Surprising Results With Codex
Hi everyone, it's me! I found an amazing example of what AI agents can really do, so let me share it with you.
OpenAI News
openai.com
What was announced?
According to OpenAI News, Asana, the project management tool company, used Codex to modernize an aging test system called Enzyme. Enzyme had gone unmaintained and was holding back efforts to modernize Asana's frontend stack.
Why it matters
Long-neglected technical debt like this tends to get harder to tackle the longer it sits, which is often why it never gets addressed. In Asana's case, the original estimate for this work was a full five years. This story is getting attention as a real-world example of how much an AI agent can compress that kind of timeline.
What changes
Using Codex, Asana finished work that was originally estimated at five years and roughly $6 million in just 1.5 weeks of engineering effort (about two calendar weeks), for an actual cost of about $12,000, roughly a 98% cost reduction.
Dive Deep
The approach: starting from a prompt of about five sentences, they ran up to four coding agents in parallel. Each agent worked on its own independent copy of the codebase, and engineers checked in on progress twice a day, reviewing every proposed change.
One interesting takeaway is that simple instructions worked better than complex configurations. Asana's CTO, Amritansh Raghav, said that not every long-running project will shrink to a few weeks, but agents can give engineers room to be resourceful.
Wrap-up
- Asana used Codex to modernize its aging "Enzyme" test system
- Originally estimated at five years and about $6 million, it was actually completed in 1.5 weeks for about $12,000
- Up to four agents ran in parallel from a roughly five-sentence prompt, with humans reviewing progress twice a day
- A great real-world reference for teams wrestling with long-standing technical debt who are considering AI agents
