Eric Glyman: Why Ramp Is Worth $60 Billion and Software Became Money

Ramp co-founder and co-CEO Eric Glyman explains why investors valued the company at $60 billion, why AI tokens turned software into a financial product, and why Ramp is building for a world where agents choose their own bank. In episode 2 of Brainfood, Simon Taylor (founder of Fintech Brainfood and head of market development at Tempo) presses Glyman on whether fintech is dead, how Ramp’s background coding agent Inspect writes most of the company’s code, what AI spend now looks like inside companies, the national trust bank charter application with the OCC, stablecoin payments, and whether Ramp will ever go public.

TLDW

Glyman argues Ramp was never really a fintech company but a productivity company, and that once AI tokens gave software a variable cost, managing software spend became a financial problem Ramp had to own. He defends the $60 billion valuation with two investor theses: a business growing over 100% a year while accelerating and generating cash, and 70% of R&D working on products nobody has seen yet. Inspect now writes 85% of Ramp’s pull requests, and non-engineers merged 7,000 of them last quarter. The conversation closes on agents: Ramp’s test is whether a perfectly informed AI agent would still choose it over every competitor.

Thoughts

Glyman’s thought experiment for the agent era is the cleanest strategic filter in the conversation: imagine an agent with perfect information about you and every competitor, and ask whether it would choose you. If the answer is no, the business is living on friction, brand and switching costs, all of which an agent ignores. What makes the test more than a slogan is the arithmetic he attaches to it. Interchange is a fixed pie that card issuers fight to split, while the median Ramp customer cut expenses by about 7 to 12% last year, a pool of value much larger than any card rebate. A company that saves customers more than it can ever give back in rewards is the one a rational agent picks. Banks that compete on points and relationships should read that as a warning, especially next to Taylor’s mention of research showing AI assistants already recommend Ramp and Mercury well above incumbent banks.

The valuation defense is more honest than most founders manage. Glyman does not pretend the multiple is cheap on trailing numbers; he says everything that grows fast looks expensive looking backward. Then he lays out the uncomfortable logic tree his co-founder Karim Atiyeh uses with investors: 70% of the R&D team works on unreleased products, so either Ramp is an incredibly bad allocator of capital or it is working on things that are hard to model. That framing invites investors to judge the team, not the spreadsheet, and it only works because the spreadsheet is already strong: growth above 100% a year, faster than a year ago, cash generation, and a card business that is still only 3 to 4% of US corporate cards. The second thesis is effectively an option on the next S-curve, priced by people who watched Ramp go from cards to bill pay to software to AI products in five years.

The Inspect story is the most practical part of the episode for anyone running a company. The headline numbers (85% of pull requests written by a background coding agent, 7,000 merged by non-engineers in a quarter) are striking, but the real point is the process that disappeared. Glyman walks through the old path for a customer bug: a support ticket, a categorized pain point, a user story, a scrum board, a prioritization meeting, a designer, an engineer, and a fix two or three months later that the customer never hears about. With Inspect, a support person or the AI itself reads the bug, ships the patch, tests it and emails the customer in 15 minutes. That is why his question, “why have you designed your organization the way you have in a world where computers can think,” lands harder than any adoption metric. Most companies are bolting AI onto workflows that only existed because coordination was expensive.

The opening argument about tokens explains why a corporate card company now runs an AI router and an internal agent harness. Traditional software had no marginal cost: using Excel ten hours a day instead of one did not raise the Microsoft bill. AI breaks that, because every token costs electricity and scarce compute, so ten times the usage can mean a ten-times larger bill. Once software has variable costs, it behaves like money and needs the same controls as any other spend. Glyman adds that AI spend already equals more than 2% of all white-collar labor spend and behaves like hiring, with agents hiring other agents and nobody knowing the final cost upfront. “I don’t sell tokens, we buy tokens” is a neat line about incentives: a router built by a buyer has every reason to cut the bill.

The closing reframe turns the “fintech is dead” provocation on its head. Glyman reads Jensen Huang’s vision, plug in a GPU, pay for the electricity, and work comes out the other end, as a description of a world where everything becomes fintech, because converting dollars into productive work becomes a core input for most businesses. If that is right, the companies that sit in the flow of funds and can actually execute (cancel a charge, move money, store cash, pay a vendor abroad on a Friday night) are better placed than the ones that only show dashboards. That is the thread linking the OCC charter, the stablecoin sandwich and the MCP server that has been live for over two years. Ramp is positioning itself less as a card issuer and more as the operational layer an agent would need to run a company’s finances.

Key Takeaways

Taylor frames the latest round as a $60 billion valuation on roughly $2 billion in revenue, about 30 times, well above Block or PayPal.
Ramp is growing more than 100% a year and more than 25% faster than it was a year earlier, at much larger scale, while generating cash.
About 70% of Ramp’s R&D team works on products that have not been released yet.
Inspect, an internal background coding agent that began as a front-end engineer’s “right-click to fix” idea, now writes 85% of Ramp’s pull requests.
Non-engineers merged 7,000 pull requests last quarter, mostly customer experience staff fixing reported bugs. Two or three years ago the number was zero.
Glyman says AI spend now exceeds 2% of white-collar labor spend and calls it the third major category of business spend, after people and vendors.
Because tokens carry real marginal cost, software has become a variable expense that needs financial controls, which is why Ramp built Router.com and Glass.
Within the next 12 months the majority of Ramp’s business is expected not to be card. Monetizable non-card volume already exceeds a quarter trillion dollars a year and is growing over 100%.
Nearly a third of customers use three or more Ramp products, and the software line is a multiple nine-figure business.
The median Ramp customer grew revenue 16% last year versus about 3 to 4% for the median US company, and cut expenses by about 7 to 12%.
Ramp serves over 85,000 businesses, the vast majority outside tech, including trash collection firms, government entities, health care providers and more than 10,000 nonprofits.
Ramp has applied for an OCC national trust bank charter, which Glyman links to businesses wanting a partner that can act on their behalf, such as moving idle cash to higher yield.
The “stablecoin sandwich” (fiat in, stablecoin in the middle, fiat out) lets Ramp pay foreign vendors after banking cutoffs and lets Latin American customers hold USDC while cards settle in local currency.
Ramp’s MCP server has been live for over two years, and agent usage has grown about 10% a week in many weeks.
Glyman expects Ramp will very likely go public one day but sees an IPO now as a distraction from keeping up with growth.

Chapters

1:49 Is Fintech Dead? Ramp as a Productivity Company

Taylor opens with his thesis that fintech is dead and Ramp is the proof, since an AI harness and a token router have little to do with payments. Glyman asks whether fintech was ever really alive, pointing out that Wells Fargo and American Express started with stagecoaches and moving paper. Companies evolve past their labels. He says he always saw Ramp as a productivity company: not about moving money from A to B, but about wasting fewer dollars.

3:30 Why AI Tokens Turned Software Into a Financial Product

Glyman describes the subtle shift that happened when tokens started to carry a cost. Classic software had no extra cost for heavier use, but AI bills scale with usage because providers pay for electricity and allocate limited supply. Going from one hour to ten hours of AI use can turn a $1,000 bill into a $10,000 one. Once software had variable cost, he says, it became money, and managing it became a natural extension of Ramp’s spend management thesis.

5:44 How Inspect Lets Non-Engineers Ship Code

Inspect is named after Chrome’s “inspect element.” A front-end engineer wanted to right-click a bug and have AI patch it, then kept extending the idea with a virtual machine, codebase access and controls. It now writes 85% of Ramp’s pull requests. Glyman says the pattern repeats across Ramp: an aggressive internal team hits the limits of available tools, builds its own, and then offers it to customers, as with Glass and, years earlier, built-in expense management.

10:26 AI as the Third Pillar of Business Spend

Classically a firm could turn capital into two things: people or vendors. Glyman says AI is now a third input, more than 2% of white-collar labor spend, behaving like people but rented, and often agentic, with one agent hiring another. Companies spending well on AI are growing and hiring faster, yet most are poorly prepared to manage it. Ramp’s model of monetizing like a card issuer while competing with software vendors such as Concur and Coupa lets it tap different profit pools, and Taylor asks whether Ramp would tell a customer it should spend more on AI. Glyman says yes, eventually.

16:35 What Justifies the $60 Billion Valuation

Glyman describes two investor camps in the round. Classic investors saw a rare financial profile: over 100% growth, accelerating, cash generating, with a tiny share of US corporate cards and under 1% of general payments. Earlier-stage investors bet on the 70% of R&D working on unreleased products. He adds that the majority of the business will soon be non-card, that software is a multiple nine-figure line, and that more customers, using more products, at larger companies, create a compounding effect.

24:20 Measuring AI Productivity and Redesigning Work

For sales the test is simply whether the team sells more. Glyman praises Ramp CPO Geoff Charles’s framework of AI capability levels and names the trap: professionals are paid to look like they know what they are doing, so few admit they cannot use the new tools. Ramp focuses on speeding up inputs rather than managing outputs. He contrasts a months-long bug fix process with a 15-minute one, and asks why organizations are still designed for a world where computers cannot think.

28:49 Competing With AI Labs for Talent

Asked how Ramp wins engineers who could join the frontier labs, Glyman says a handful of people a year are wired for novel AI research and should go there. Many other brilliant people want to see their work change lives, and Ramp offers a lever: customers growing revenue 16% against a 3 to 4% national median. He also pitches a culture of founders, small teams, monthly launches and early access to tools, noting Ramp was among Cursor’s first ten customers and that Cognition filmed its Devin launch video in Ramp’s office.

33:17 Customers Beyond Tech and Partners Over Zero-Sum

The vast majority of Ramp’s 85,000-plus customers are not tech companies, and Glyman describes the moment a small organization first sees an expense handle itself as stepping into the future. On partners such as Stripe and Marqeta, he rejects zero-sum thinking, citing Microsoft and Apple, and Netflix as a major AWS customer while competing with Amazon’s studios. Stripe’s purchase of OpenRouter on the same day Ramp launched Router.com did not bother him, he says, because both businesses are small relative to the world.

36:15 The OCC Trust Charter and Stablecoin Payments

Ramp has applied for a national trust bank charter, and Glyman credits the current administration with recognizing that the post-2008 worry about too few, too-big banks has only grown. He sees demand for a partner that can act within a business’s stated investment policy. On stablecoins, he says crypto lost the plot by pitching the math instead of the benefit, then describes real use: paying a foreign vendor after Friday’s cutoff, and letting Latin American employees swipe cards in local currency while the company holds USDC.

43:18 Going Public, Agents and Why Everything Becomes Fintech

Glyman expects Ramp to go public eventually but says the effort would distract from growth for now. On agents, he thinks every outcome is possible, including agents running all of a company’s operations, and Ramp’s MCP server usage is growing fast. He closes by reading Jensen Huang’s vision of GPUs turning electricity into work as a world where all software and work becomes fintech, and says Ramp’s mission stays the same: more from every dollar, hour, and now token.

Notable Quotes

“Once software started to have marginal costs in variable cost, I think software in some sense became money. It actually became a financial product.”
Eric Glyman, on why AI tokens pulled Ramp into software spend

“Everything that grows fast looks incredibly expensive and everything that grows slow looks very cheap.”
Eric Glyman, on judging Ramp’s valuation by trailing numbers

“Either we are incredibly bad allocators of capital and we’re just burning money or we might be working on things that are hard to model.”
Eric Glyman, on 70% of R&D building unreleased products

“Why have you designed your organization the way you have in a world where computers can think?”
Eric Glyman, on the question AI should force every company to ask

“The idea that someone has margin, you need to go and take it as I think it’s just a very old way of thinking.”
Eric Glyman, on partnering with Stripe and other infrastructure providers

“Let’s say an agent with perfect information about you and all your competitors. The question is, would that agent choose to work with you or not? And if the answer is no, in the long run, you might have a problem.”
Eric Glyman, on the test Ramp designs against

“People always want a little bit more for less. I think that is certainly going to be true in two years. I think it’s going to be true in 100 years.”
Eric Glyman, on what will not change about Ramp’s mission

Watch the full conversation here.