AI News
All stories
Reported

Ramp says its Inspect coding agent now writes about 75% of its code

Ramp engineer Zach Bruggeman says Ramp's internal Inspect coding agent writes and ships about 75% of its code, and that Inspect is now available outside Ramp via a waitlist.

AI in Practice · · 3 min read

Ramp says its internal coding agent, Inspect, now writes and ships about 75% of the company's code. Engineer Zach Bruggeman made the claim in a talk at CODE Summit 2026, as reported by AI Engineer. Bruggeman also said Inspect is now available outside Ramp, and interested teams can join a waitlist at inspect.com/aie. The talk offers no independent figures for the 75% number.

Bruggeman described Inspect as a collaborative coding agent. Each session runs remotely and is multi-user, so a developer can work with teammates, Slack, Linear and other integrations inside the same session. Inspect is also wired into Ramp's internal tools, including telemetry and databases, so any employee's session starts with access to the company's data. Bruggeman said the team bet about a year ago on building its own agent, and that the technology is no longer the hard part. He said the sandboxes now start in less than 5 seconds and that the team has moved its focus to culture.

Ramp runs one shared Inspect for the whole company. Bruggeman argued that setup time adds up. At an organization of a few hundred or a thousand people, 15 minutes of configuration per person becomes days of routine work, and knowledge of who has the best settings is unevenly distributed. With one shared version, a pattern that works for one engineer can be rolled out to everyone, and most improvements are invisible to the people using it. Bruggeman also said the team found it did not need formal training on how to use AI well. Because sessions are public by default inside the company, staff can watch colleagues work in channels they already share. They can also ask Inspect to review their own recent sessions to see how they structured prompts.

Bruggeman tied trust to visibility. Each change request can carry the session history, the tools Inspect called, before and after screenshots, and in some cases a video of a verification agent clicking through the app in a browser. He contrasted this with receiving a 400-line diff and hoping it works. The talk includes a debugging episode from February that he presented as an example. An engineer, Andrew, reported a bug in Inspect, and Inspect blamed the toolkit used to debug it. Bruggeman and a colleague, Dylan, were not convinced and pressed the agent for a more specific cause. Inspect then pointed to a telemetry library that calls fetch but never reads or discards the response body, a limitation Bruggeman said was specific to the Cloudflare platform at the time. Inspect produced a patch that explicitly cancels the response body, and Bruggeman said there had been no errors since deployment. The example is anecdotal. It illustrates his point: the debugging happened in a shared thread, so Andrew became involved rather than receiving a result afterward.

Bruggeman also credited the Review Buddy code review agent to two engineers, Daniel and Pranjal, who built it on top of Inspect's automation triggers, including triggers that fire when a pull request is opened. Ramp had previously used off-the-shelf verification products and alternated between them. Bruggeman presented the project as evidence that employees could start useful tools without asking permission. He said it made the Inspect team think of the product as more than a coding agent. The talk gives no comparison of Review Buddy's results with the earlier tools, and the 75% figure is not broken down by team or by type of change.

What changed

Ramp's internal Inspect coding agent is now available outside Ramp via a waitlist at inspect.com/aie.

Type: New capability · Applies to: Inspect

Why it matters

The claim gives a concrete internal benchmark for how much of a production codebase an agent can write, though it comes from the company itself and is unverified. Teams weighing shared coding agents may find the emphasis on public sessions and reviewable verification relevant.

What you can do

Engineering leaders evaluating coding agents can request access to Inspect via the waitlist and test whether shared, reviewable sessions suit their workflow.

Original source

We have no link to the original source for this story yet.

Where we saw it

1 source

AI Engineer

AI wrote this summary from the sources listed, and nobody reviewed it individually before publication. Check the original source before you rely on it. How it works