Asana clears 5 years of engineering work in 2 weeks with Codex

Asana's migration off Enzyme highlights a shift in how engineering organizations can approach legacy technical debt. The old testing system was no longer actively maintained and had become a blocker to modernizing the frontend stack. By using OpenAI Codex, Asana's engineers reduced a projected five-year, $6 million staffing effort to about two weeks and $12,000 in direct costs, suggesting that AI agents can compress timelines dramatically when tasks are well-scoped.

The project's success appears to hinge on two factors. First, simple instructions outperformed elaborate setups, and having an engineer check progress twice daily kept the agents on track. Second, human review of each proposed change remained essential, indicating that agents augment rather than replace engineering oversight. Asana's CTO notes that not every years-long project will collapse into weeks, but the experience likely expands the set of migrations and rewrites the company now views as feasible.

Looking ahead, the real test is whether this approach scales to other long-running software problems Asana previously avoided. If the pattern holds, other companies may revisit their own technical debt backlogs, though outcomes could vary depending on codebase complexity and the quality of agent instructions.