The PlaybookAI Strategy

Ramp Is Worth $60B. It built its own code reviewer in two weeks.

Eric Glyman announced $1.85B at a $60B valuation. Ramp's engineers built their own code reviewer in two weeks rather than buy one. Here is why.

John TanOctober 15, 20264 min read

On October 15 Eric Glyman announced that Ramp had raised $1.85B at a $60B valuation. A company that size can buy any developer tool on the market. Its engineers wrote their own instead, in two weeks, and it now writes most of the code.

Eric Glyman
Eric
Glyman

"Today, Ramp raised $1.85B at a $60B valuation. Seven years ago, Karim and I wrote our mission on a Post-it: 'Save companies time and money.' Our mission hasn't changed. It's just gotten bigger."

Eric Glyman  ·  X  ·  October 15, 2026

The raise is the headline. The interesting part is what a company at that valuation decided it could not buy.

$1.85B

Raised at a $60B valuation, announced by the CEO on October 15, 2026.

2 weeks

Build time for Inspect, their in-house coding agent, by three engineers.

~1,600

People at the company when it was built. Not a two-person startup improvising.

The buy decision they did not make

Every vendor in the category would have taken Ramp's call. At $60B the procurement cost of a code review tool rounds to nothing, and the engineering time to build one is the expensive part. The standard logic says buy.

They built. Inspect took two weeks and three engineers, and the reasoning its lead gave is the part worth copying. Not that the vendors were bad. That the integration was the product.

Figure 1. What the build-versus-buy answer actually turns on
The tool is cheap. The context is not for sale.
A TOOL YOU BUY A TOOL YOU BUILD The repository The pull request Nothing below this line. The repository The pull request Six years of specs, requests and roadmaps The log schema and the one API key it needs The model is the same in both columns. It is the same model you can rent. The difference is everything the company already knew about itself. That is the part no vendor can ship you, at any price.
Reach compiled from Ramp's own account of Inspect, published by Linear, captured 2 November 2026. Figure by nativefirst.

Ramp had six years of specs, customer requests and roadmaps sitting in one system, reachable by an agent. That let Inspect trace a thread from a customer request to a product spec to the code that implements it. A bought tool sees the diff. Theirs saw why the diff exists.

That is not a story about Linear, and it is not a story about two weeks. It is a story about what happens when the institutional memory is already machine readable on the day you decide to build. The two weeks is the visible part. The six years is the moat.

What $60B buys that a seed company does not have

It is tempting to read this as "big company, big budget." The budget is the least of it. Three things were in place, and only one of them costs money.

The history was already queryable. Not archived. Not in a wiki nobody opens. Reachable through an API, by a machine, without anybody rekeying it. Most companies at any size fail this one and it is the whole precondition.

A central team set what agents can touch. Ramp's own framing is that a central team defines what agents can access and do, so functional teams build their own automations on the same rails. That is one decision made once, and it is what lets the second and fifth agent take two weeks instead of two quarters.

They knew their own plumbing. The engineer who led it put it plainly: internally they know the one API key it needs and precisely what the schema of their logs looks like. A vendor has to discover that for every customer. You already have it, which is an advantage you are probably not counting.

When buying is still the right call

Most companies reading this should buy, and the reason is the figure above, not a lack of ambition. If your last six years of product decisions live in Slack threads, closed tickets and three people's heads, an in-house agent inherits none of it. You would be building a worse version of a tool you can rent, and paying engineering time for the privilege.

Worth saying plainly about the result too: the reported number for Inspect is 75% of merged pull requests, which is a count of volume and not a measure of how hard the work in them was. It dates from September. It is still a remarkable number for three engineers and two weeks, and it is not the same claim as "the agent does three quarters of the engineering."

The bottleneck moved, and that is the real result

Ramp's own account says the speed moved their bottleneck from writing code to reviewing it. Read that twice, because it is the thing most teams get wrong about this whole category. The win was not less work. The win was work arriving at a different place.

If you ship an agent that writes code and you have not also decided who reviews its output and how fast, you have not removed a constraint. You have moved it onto whoever is closest. That person is usually one named human, and the queue in front of them is now longer than it was.

What's next

  1. Drop the build-or-buy framing. Ask instead what an internal agent could reach if you started it this month.
  2. List what a machine can query today. If the answer is the repository and nothing else, buy.
  3. Put your own decisions somewhere queryable. Six years of them is the thing no vendor sells you.

Context is the moat.

Portrait of John Tan
John Tan ↗

Founder of nativefirst.ai. Runs everything on agents and writes down what works.

Sources
1Eric Glyman on X, 15 October 2026: "Today, Ramp raised $1.85B at a $60B valuation. Seven years ago, Karim and I wrote our mission on a Post-it: 'Save companies time and money.' Our mission hasn't changed. It's just gotten bigger." The CEO's own announcement, not secondary reporting.
2Ramp's account of Inspect, its in-house background coding agent, built in two weeks by Zach Bruggeman with Jason Quense and Rahul Sengottuvelu at a company of roughly 1,600 people, published by Linear at linear.app/customers/ramp. The claim that it writes 75% of merged pull requests is reported there and dates from September 2026; it is a volume measure, not a measure of the difficulty of the work, and this post does not restate it as a current figure.
3Zach Bruggeman on build versus buy: "It's a really tight integration with our development lifecycle and our tooling", with the team knowing internally "the one API key it needs and precisely what the schema of their logs looks like". Cristina Cordova on governance: "A central team defines what agents can access and do, so functional teams can build their own automations safely on the same rails." Both via linear.app/customers/ramp.
4"The speed has moved Ramp's bottleneck from writing code to reviewing it" is Ramp's own characterisation as published by Linear, not an independent measurement.

Read next

AI is moving really fast. Keep up.

Read the Playbook