Asana’s engineering team removed the deprecated Enzyme testing framework from its codebase in roughly two calendar weeks by using OpenAI Codex. Model and infrastructure expenses for the effort were about $12,000, compared with an earlier staffing-based estimate of roughly $6 million and a timeline of about five years.
Why the migration was needed
Enzyme had fallen out of active maintenance and was becoming a barrier to modernizing Asana’s frontend stack. The outdated testing tool made upgrades and future development more difficult, so the company decided a migration was necessary.
How Codex was used
The team employed coding agents powered by OpenAI Codex. Work began from a prompt composed of five sentences; up to four agents worked in parallel, each operating in its own copy of the codebase. An engineer checked progress twice daily and reviewed every proposed change before approval. The team found that simpler, clearer instructions produced better results than more elaborate configurations.
Time and cost comparison
The engineering effort amounted to about 1.5 weeks of work distributed over two calendar weeks. Model and infrastructure costs totaled approximately $12,000. By contrast, the prior plan estimated the same work would take at least five years and cost roughly $6 million under the previous staffing assumptions.
Implications for future projects
With the migration complete, Asana now plans to test agents on other migrations, rewrites, and performance problems it previously assumed would require years. The company aims to use automation to free engineers from repetitive tasks while keeping humans responsible for review and approval of changes.
Conclusion
Asana’s experience indicates that AI-assisted, human-supervised code changes can make large modernization efforts feasible in far shorter timeframes and at much lower cost than earlier projections suggested.
