Asana uses OpenAI Codex to clear 5 years of legacy code fixes in 2 weeks for tens of thousands of dollars
Task management platform Asana shares its success in using AI to transform legacy software testing systems, saving massive amounts of time and costs.
Asana has revealed a case study on using the OpenAI Codex model to tackle major technical debt, successfully migrating its frontend test suite from the deprecated Enzyme library to React Testing Library (RTL).
Originally, Asana's engineering team estimated this task would accumulate 5 years of work, translating to roughly $6 million USD in personnel costs. However, by leveraging OpenAI Codex, they completed the project in about 1.5 weeks of actual work (within a 2-week calendar window) while spending only about $12,000 USD on API and AI infrastructure costs.
Behind this success, the engineering team used short prompts of about 5 sentences to instruct Codex to run up to 4 AI agents in parallel across different code folders, with human engineers reviewing the code twice a day before merging it into the main system.
Notably, the 5-year figure represents an estimate of accumulated backlog work rather than core feature development, highlighting AI's potential to handle repetitive, time-consuming mechanical tasks in the software industry.
This demonstrates that current AI models can help organizations resolve multi-year software backlogs in a matter of weeks at a drastically lower cost, potentially reshaping how IT budgets and project management are approached in the future.