Asana uses AI to slash engineering time

Asana, a leading work management platform, has achieved a remarkable feat in software development, completing five years' worth of engineering work in just two weeks. The company utilized OpenAI Codex, powered by advanced AI models, to overhaul its testing system, removing an outdated framework called Enzyme. This project, expected to take half a decade with traditional methods, was finished for a mere $12,000 in model and infrastructure costs. This figure starkly contrasts with Asana's initial estimate of approximately $6 million for staffing the same project.

The outdated Enzyme testing system had become a significant impediment to modernizing Asana's frontend stack. Recognizing this, Asana's engineering team turned to Codex, providing it with a concise five-sentence prompt. The AI then generated code changes, with up to four coding agents working in parallel on separate copies of the codebase. An engineer supervised the process, reviewing proposed changes twice daily. This efficient workflow led to the complete removal of Enzyme in about 1.5 weeks of focused engineering effort, spread across two calendar weeks.

A New Horizon for Engineering Projects

This success has fundamentally altered Asana's perception of what long-running software projects are feasible. The company now sees potential for AI agents to tackle other complex migrations, rewrites, and performance optimizations that were previously deemed too time-consuming or costly. Asana CTO Amritansh Raghav noted, "Not every years-long project will collapse into weeks. But agents can give engineers more room for craft, and make once-impossible work worth attempting." The goal is to free up engineers to concentrate on higher-level design and innovation, while AI handles the more labor-intensive coding tasks.

The implications of this demonstration are profound for enterprise software development. For years, companies have grappled with technical debt and the slow, expensive process of updating legacy systems. Asana's experience suggests that AI-powered coding assistants can dramatically accelerate this modernization, potentially unlocking significant cost savings and faster innovation cycles. This approach aligns with broader trends in the AI industry, where tools like Codex are moving beyond simple code completion to more complex autonomous development tasks.

While Asana is a major player in the work management space, with StartupHub.ai data giving it a score of 75/100, its adoption of AI tools like Codex is an indicator of how AI is reshaping engineering practices across the board. Competitors such as Linear (score 58/100) and Motion app (score 64/100) are also vying for market share, often by emphasizing streamlined user experiences and productivity gains, but Asana's move highlights a deeper technological shift.

The Economics of AI-Assisted Development

The economic disparity between the AI-assisted approach and traditional staffing is staggering. While Asana estimated $6 million for human engineers to complete the Enzyme migration, the AI solution cost around $12,000 for model usage and infrastructure. This 500x cost reduction, achieved in a fraction of the time, points to a future where AI dramatically alters the economics of software development. It's not just about speed; it's about making previously prohibitively expensive projects financially viable.

This development also raises questions about the future of developer roles. While Asana emphasizes that engineers still review and approve changes, the nature of their work shifts. Instead of spending years on a single migration, they can oversee AI agents, focusing on strategic oversight and complex problem-solving. This could lead to a more fulfilling and impactful engineering career, but also necessitates new skill sets in prompt engineering and AI system supervision.

The specific details of Asana's prompt and the exact frontier models powering Codex for this task are not elaborated upon in the announcement. Understanding the nuances of the prompts that yield such transformative results will be key for other organizations looking to replicate this success. However, the core message is clear: AI agents are no longer just theoretical tools; they are capable of delivering tangible, large-scale results for enterprise engineering teams.