How Asana Used OpenAI Codex to Complete a 5-Year Software Migration in 2 Weeks
Article Summary
Asana successfully completed a significant software modernization project in approximately two calendar weeks, a task previously estimated to require five years of engineering effort. The project involved the removal of Enzyme, an outdated testing system that had fallen out of active maintenance, posing a significant blocker to upgrading Asana's code and modernizing its frontend stack.
To achieve this acceleration, Asana leveraged OpenAI Codex, powered by frontier models, within its engineering organization. The process began with a concise five-sentence prompt, which directed up to four coding agents to work in parallel on separate copies of the codebase. An Asana engineer maintained oversight, checking progress twice daily and reviewing every proposed change, noting that simpler instructions yielded better results than more elaborate setups.
The material metrics for this undertaking are notable: the total engineering effort spanned 1.5 weeks across two calendar weeks. The combined model and infrastructure costs for using Codex amounted to $12,000, a dramatic reduction compared to the $6 million and five-year timeline projected for the previous staffing plan. This success has altered Asana's perception of the feasibility of long-running software projects. The company now plans to apply this agent-driven approach to other complex tasks, such as migrations, rewrites, and performance problems, which were previously considered multi-year endeavors.
Asana's Chief Technology Officer, Amritansh Raghav, indicated that while not all extensive projects will condense into weeks, this method empowers engineers with more creative space by making once-impossible work attainable. The goal is to allow engineers to focus on higher-value, craft-oriented tasks, with human review remaining a critical component of the development process.
To achieve this acceleration, Asana leveraged OpenAI Codex, powered by frontier models, within its engineering organization. The process began with a concise five-sentence prompt, which directed up to four coding agents to work in parallel on separate copies of the codebase. An Asana engineer maintained oversight, checking progress twice daily and reviewing every proposed change, noting that simpler instructions yielded better results than more elaborate setups.
The material metrics for this undertaking are notable: the total engineering effort spanned 1.5 weeks across two calendar weeks. The combined model and infrastructure costs for using Codex amounted to $12,000, a dramatic reduction compared to the $6 million and five-year timeline projected for the previous staffing plan. This success has altered Asana's perception of the feasibility of long-running software projects. The company now plans to apply this agent-driven approach to other complex tasks, such as migrations, rewrites, and performance problems, which were previously considered multi-year endeavors.
Asana's Chief Technology Officer, Amritansh Raghav, indicated that while not all extensive projects will condense into weeks, this method empowers engineers with more creative space by making once-impossible work attainable. The goal is to allow engineers to focus on higher-value, craft-oriented tasks, with human review remaining a critical component of the development process.
AI Analysis
## Core Assessment
Asana dramatically reduced a five-year, $6 million software modernization project to approximately 1.5 weeks of engineering effort and $12,000 in model and infrastructure costs by leveraging OpenAI Codex-powered coding agents. The task involved the systematic removal of Enzyme, an outdated testing system that had become a blocker to upgrading Asana's frontend stack. This success signals that AI-powered coding agents can provide immense leverage for well-defined, systematic code migrations or refactoring tasks. The process still required significant human oversight: an Asana engineer checked progress twice daily and reviewed every proposed change. Simpler instructions for the agents proved more effective than complex setups, suggesting prompt engineering remains a key skill. The practical consequence for Asana is a re-evaluation of what types of long-running, code-heavy projects are now considered feasible, shifting the bottleneck from manual, repetitive coding to verification and strategic task identification. It seems "impossible" just means "too expensive the old way" for some projects now.
## Impact and Advice
* **Software Architect**: Evaluate existing technical debt involving outdated libraries or frameworks that require systematic code changes. Prioritize candidates for AI-assisted migration, focusing on projects with clear, measurable outcomes and defining simple, explicit prompts for coding agents.
* **Engineering Lead**: Pilot AI-powered coding agents for specific, well-bounded code modernization efforts within your teams. Implement a rigorous human-in-the-loop review process, requiring engineers to validate every proposed change, similar to Asana’s approach.
* **Product Manager**: Reconsider the feasibility and timelines for code migration, rewrite, or performance projects previously shelved due to estimated multi-year timelines or high costs, recognizing that these tools primarily address existing code cleanup rather than new feature development.
* **IT Operations Lead**: Factor in and monitor API and infrastructure costs associated with large language models and coding agents when budgeting for major code transformations, noting the $12,000 cost incurred for Asana's project.
## Bottom Line
This case demonstrates a significant acceleration in tackling specific, large-scale code modernization tasks using AI coding agents, dramatically cutting both time and cost. The key takeaway is the shift in engineering effort from manual execution to oversight and strategic task definition, with human review remaining a non-negotiable component.
Asana dramatically reduced a five-year, $6 million software modernization project to approximately 1.5 weeks of engineering effort and $12,000 in model and infrastructure costs by leveraging OpenAI Codex-powered coding agents. The task involved the systematic removal of Enzyme, an outdated testing system that had become a blocker to upgrading Asana's frontend stack. This success signals that AI-powered coding agents can provide immense leverage for well-defined, systematic code migrations or refactoring tasks. The process still required significant human oversight: an Asana engineer checked progress twice daily and reviewed every proposed change. Simpler instructions for the agents proved more effective than complex setups, suggesting prompt engineering remains a key skill. The practical consequence for Asana is a re-evaluation of what types of long-running, code-heavy projects are now considered feasible, shifting the bottleneck from manual, repetitive coding to verification and strategic task identification. It seems "impossible" just means "too expensive the old way" for some projects now.
## Impact and Advice
* **Software Architect**: Evaluate existing technical debt involving outdated libraries or frameworks that require systematic code changes. Prioritize candidates for AI-assisted migration, focusing on projects with clear, measurable outcomes and defining simple, explicit prompts for coding agents.
* **Engineering Lead**: Pilot AI-powered coding agents for specific, well-bounded code modernization efforts within your teams. Implement a rigorous human-in-the-loop review process, requiring engineers to validate every proposed change, similar to Asana’s approach.
* **Product Manager**: Reconsider the feasibility and timelines for code migration, rewrite, or performance projects previously shelved due to estimated multi-year timelines or high costs, recognizing that these tools primarily address existing code cleanup rather than new feature development.
* **IT Operations Lead**: Factor in and monitor API and infrastructure costs associated with large language models and coding agents when budgeting for major code transformations, noting the $12,000 cost incurred for Asana's project.
## Bottom Line
This case demonstrates a significant acceleration in tackling specific, large-scale code modernization tasks using AI coding agents, dramatically cutting both time and cost. The key takeaway is the shift in engineering effort from manual execution to oversight and strategic task definition, with human review remaining a non-negotiable component.