Blog/customersContact sales

Navan's most recent migration touched 8 applications and 15 libraries, and finished in about a week without freezing development. Two years ago, a comparable migration took over a month, and a bad merge locked the master branch for a day. The difference is Cursor's Cloud Agents, which let Navan plan the migration once and run it across 8 applications in parallel on its own self-hosted machines, instead of one engineer working through it alone.

An 8-app migration finishes in a week instead of a month

Two years ago, moving Navan's codebase to a new Spring version took Rahul Popuri, Sr. Staff Software Engineer, over a month. Incomplete pushes and reverts locked the master branch for a day during that stretch. The disruption didn't end when the lock lifted: other engineers were still catching up on landing changes days later. The bigger cost wasn't the lock itself. It was the month of an engineer's time spent on the migration instead of shipping features or fixing bugs.

Navan's most recent migration, to Spring Boot 4, finished in about a week, with less disruption to ongoing development. With Cursor, Rahul planned the upgrade in advance, then delegated each component to Cloud Agents running on Navan's self-hosted machines, so 8 applications and 15 libraries moved at the same time instead of one after another. The agents used Navan's internal tooling and spun up browsers on those same machines to click through flows and verify their changes.

In the past, one bad merge brought workstreams to a standstill. This time I had eight apps moving at once while minimizing any disruption to ongoing development.

Rahul PopuriSr. Staff Software Engineer, Navan

Rahul wasn't watching a single job run. He was monitoring multiple agents upgrading different applications at once, a scale of parallel work he says wouldn't have been possible even a few months earlier.

Cloud Agents run the real workload

Rahul moved almost all of his work onto Cursor's Cloud Agents. GitHub work runs on Cursor-managed infrastructure. Internal MCPs for Snowflake and Slack run on Navan's own self-hosted machines, where agents can also spin up a browser to verify their own changes. Automations, Skills, and Memories run on top of that setup, so eval loops and migrations continue asynchronously without someone watching them.

That pattern has since scaled well past one engineer: roughly 200 people at Navan now actively use Cursor's Cloud Agents, a 10x increase since April 2026.

Before Cursor, Navan's engineers had built real infrastructure around their previous coding agent, but it required someone at one desktop with the session open the whole time. That ruled out the parallel, asynchronous work Navan needed.

The same leverage shows up on product teams

On the product side, Elton Yang, Staff Software Engineer on expense and billing, builds new agents for expense and billing workflows starting from a plain description. Cursor follows the team's existing design patterns and writes the agent's prompts itself, which means testing starts the same day instead of after hours spent writing the initial prompt by hand.

We don't even have to write out the agent prompt ourselves anymore. Cursor follows our prior designs and gets an agent working the same day.

Elton YangStaff Software Engineer, Navan

That shift changed what a small team can carry. After many engineers moved off Navan's dedicated AI-agent team, one engineer took over feature work that previously needed 2 or 3 engineers across multiple sprints. One of the clearest examples is manual bank linking: the one engineer recreated an Angular form in React and built the backend to connect it, moving from backend-only into full-stack work when the task required it. The individually owned workstream stayed focused on new features instead of sliding into pure maintenance.

Cursor also changed what covering for a teammate looks like. When Elton had to take over two codebases he did not know well while a colleague was on leave, Cursor handled the investigation for him, pulling logs and database records and laying out the issue before Elton opened a single file.

I wasn't too familiar with either codebase. I said, 'Here is the issue, help me figure this out.' It pulled the logs and the database and walked me through it like I had been living in that repo for months.

Elton YangStaff Software Engineer, Navan

This saved roughly two days of ramp-up time across the two codebases, plus the stress of learning them cold.

Review still rests with people. Elton treats the agent as a pair programmer, not a coding oracle: he watches the agent's reasoning as it runs and corrects wrong assumptions as soon as he sees them. Once code is generated, it goes through the same human and AI review, testing, and staging process as anything else, and complex changes are verified in staging before production. Agents plan and execute the work, but engineers make the judgment calls and own the final merge.

---

If you're running migrations or parallel workstreams that used to need a full team and a locked codebase, reach out to start a Cursor trial.