Asana just pulled off something that should make every engineering leader sit up: a migration the company estimated would take five years and cost around $6 million in staffing was finished in two calendar weeks for about $12,000 in model and infrastructure costs. The project? Ripping out Enzyme, a JavaScript testing utility that had fallen out of active maintenance and was actively blocking Asana from modernizing its frontend stack.
The mechanics were surprisingly straightforward. Engineers fed Codex a five-sentence prompt, and up to four coding agents spun up in parallel, each working in its own copy of the codebase. A human engineer checked progress twice daily and reviewed every proposed change before anything merged. Asana’s team discovered that simpler instructions outperformed a more elaborate setup—a pattern that echoes what many teams have found with agentic coding: clear, constrained prompts beat sprawling specification documents.
The cost comparison is the headline number: roughly $12,000 versus an estimated $6 million under the old staffing plan, a 500x difference. But the more interesting shift is psychological. Asana’s engineering leadership said the experience changed which long-running software projects the company believes are practical to take on. Migrations, rewrites, and performance problems that once sat in the “someday maybe” pile now look tractable.
There’s a caveat worth keeping in mind: Enzyme removal, while tedious and sprawling, is also a relatively mechanical task with clear success criteria. Codex didn’t design a new architecture or make judgment calls about tradeoffs—it executed a well-understood removal at scale. The human review layer remained essential, and Asana’s own writeup is careful not to overclaim. “Not every years-long project will collapse into weeks,” the company noted, but added that agents can give engineers “more room for craft—and make once-impossible work worth attempting.” That framing—augmentation rather than replacement—seems to be where serious engineering organizations are actually landing with these tools.
