Asana turned a five-year, six-million-dollar migration into a two-week, twelve-thousand-dollar project with one scoping pattern, and that same pattern is now sitting inside a free tool you already have access to.

By the end of this you should be able to take the internal-tool idea your team has been sitting on since spring, the one that never made it onto the engineering roadmap, and scope it using the exact pattern that just let Asana turn a five-year infrastructure migration into a two-week project. Not because your idea is that big. Because the thing that made a $6 million project cost $12,000 is a scoping habit, not an engineering budget, and you already have access to the tools that run it.

Asana's engineering team had assumed for years that a specific migration was a five-year problem. Last week we found out it wasn't. It was a scoping problem.

The pattern, not the magic prompt

Vibecoding gets sold as one giant prompt: describe the app, hit enter, get software. That is not what happened at Asana, and it is not how this works once the job is real. What actually worked was three things stacked together. A problem small enough to describe in five sentences. A finish line that does not require anyone's judgment call to recognize as reached. And a person checking in on a fixed schedule instead of waiting around for a big reveal. Call it bounded scope, a clear definition of done, and a checkpoint cadence. None of that is a technical skill. It is closer to how you'd brief a contractor, and it works because it gives an agent a fence to work inside instead of an open field.

What Asana actually did

Asana's problem was Enzyme, an old testing framework that had fallen out of maintenance and was blocking the company from modernizing its frontend stack. The internal staffing estimate to remove and replace it was five years and roughly $6 million, according to OpenAI's case study on the migration. That is the kind of number that makes a project quietly disappear from the roadmap forever, which is exactly what had happened to it.

Instead, Asana engineers wrote a five-sentence prompt describing the removal, then let up to four instances of OpenAI Codex work in parallel, each in its own separate copy of the codebase. An engineer checked progress twice a day and reviewed every proposed change before it merged. Total engineering effort: about a week and a half, spread across two calendar weeks. Total model and infrastructure cost: about $12,000. Asana's CTO, Amritansh Raghav, put it plainly: "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 detail worth sitting with is the one Asana called out about itself: simpler instructions worked better than a more elaborate setup. Five sentences, not a fifty-page spec. The narrow scope did the work that a longer prompt would have tried and failed to do, because a longer prompt is usually somebody trying to anticipate every branch instead of defining what done looks like and then showing up twice a day to check.

This is not a one-company fluke. On Codex's own product page, Sierra says it now ships in a weekend what used to take a quarter, and Harvey reports early iteration time dropped 30 to 50 percent, freeing its engineers for higher-leverage decisions instead of setup work. Different companies, different codebases, same shape underneath: bound the problem, define done, check in on a cadence, and let the agent run in the gaps between check-ins.

Why this isn't just an engineering-org story

Here is the part that matters if you don't run an engineering team. Asana could attempt this because coding agents got more capable and radically cheaper at the same time, and that same economics just landed on a tool non-engineers already use. Replit announced Free Mode this week, powered by GPT-5.6 Luna, letting anyone plan, shape, and build software without burning through paid usage while they're still figuring out what they want. Replit co-founder Amjad Masad said the recent OpenAI price cuts were what finally made offering that experience to millions of users, for free, possible. Sam Altman framed the stakes even bigger: get software creation cheap enough and you get "another renaissance-level entrepreneurial boom like we've never seen before."

That is the same underlying shift as the Asana story, aimed at a different audience. Asana used cheap, capable agents to attempt a project too expensive to staff. You can use the same cheap, capable agents, on a tool with a genuinely free tier, to attempt the internal tool your team already agreed was worth building and never scheduled.

Where this goes wrong

The mistake people make when they hear "five-sentence prompt" is assuming brevity is the trick, so they write five vague sentences and wait for a miracle. It isn't brevity, it's boundedness. "Remove this specific deprecated testing library from this specific codebase" fits in five sentences because the scope was already narrow enough to fit in five sentences. "Build me a CRM" is not narrow, and shrinking the wording doesn't fix that. If your project needs a paragraph to explain what problem it solves and a full sentence to explain how you'll know it's finished, that paragraph and that sentence are your real scope. Write the prompt at that size instead of forcing it smaller for its own sake, and keep the twice-a-day check-in regardless of how confident the agent's first output looks. The check-in is not a formality. It is the only thing standing between "the agent is on track" and "the agent quietly wandered off scope four hours ago."

Try this today

Take the internal-tool idea that has been sitting in a Slack thread since spring. Write down the problem in five sentences or fewer, including what "done" looks like in a way you could check without asking anyone's opinion. Open Replit's Free Mode and hand it that exact scope, then look hard at what it proposes before you build anything further. Set two check-ins for yourself, today and tomorrow, instead of waiting for a finished product to review cold. If the scope won't fit in five sentences, that's your answer: it isn't this week's project yet.

Asana didn't clear five years of engineering work in two weeks because the agents got smarter overnight. It cleared it because somebody finally wrote down what the project actually was, in five sentences, instead of the five-year version everyone had quietly agreed to believe.