Y Combinator4 min read

Ramp's Eric Glyman: AI harness Glass drives 25% faster growth at $2B revenue run-rate

Ramp's co-CEO explains how an internal AI "company brain" called Glass helped the fintech accelerate growth and cut customer expenses by 7.5% a year.

AI summary of “How To Build The Future: Eric Glyman, Co-CEO of Ramp”

Key takeaways

  • Eric Glyman says Ramp now does over $2 billion in annual revenue, growing more than 25% faster than a year ago.
  • The median Ramp customer saves 7.5% a year on expenses, up from about 5% a year ago.
  • Ramp paused hiring for a full quarter earlier this year to force deeper AI adoption instead of hiring around gaps.
  • Glyman says non-engineers created 7,000 pull requests in the second quarter using Ramp's internal coding agent, Inspect.
  • He argues "Jevons paradox broke" over the last two months as cheaper intelligence has not increased usage proportionally.

From a refund bot called Paribus to founding Ramp

Eric Glyman, co-founder and co-CEO of Ramp, told Garry Tan that he and co-founder Kareem built Paribus, an early automated agent that scanned Gmail or Yahoo inboxes for receipts and emailed retailers for refunds when prices dropped. The idea came from a personal frustration: a flight price drop that let him and friends claim back almost $1,000. Paribus launched through YC (after being rejected once and reapplying) and grew to nearly a million daily users before being acquired by Capital One.

Early growth came from scrappy tactics, including a Reddit post on the "Beer Money" subreddit. Glyman said a key fix was discovering, through informal customer research, that people were wary of Paribus keeping 25% of savings. Introducing a two-sided referral deal, where both referrer and referee got a lower 20% rate, made growth "explode."

Time at Capital One shaped Ramp's founding insight: credit card point and cashback programs were often structured so companies profited by encouraging more spending rather than helping customers spend less. Ramp was built instead to compete on "more dollars in your account," using profit pools from financial products to fund better software.

Ramp's "end of the beginning" at $60 billion

Glyman said Ramp has reached a $60 billion valuation and described this as "the end of the beginning." The first chapter was building a unified system spanning cards, expenses, bills, procurement and travel, replacing the fragmented tools most companies use. He said the median company on Ramp now saves 7.5% per year on expenses.

The next chapter, he said, shifts focus from cutting waste to accelerating revenue growth, including a new accounts receivable product and AI tools like Glass and Inspect aimed at making every employee more productive rather than just reducing costs.

The top 1% of AI spenders, and a hiring freeze

Glyman cited a Ramp statistic that the top 1% of spenders account for 80% of spend on AI tokens, calling the uneven adoption of AI "a societal, GDP-affecting" issue. Inside Ramp, he said this same pattern created internal strain: some engineers became "100x" productive while designers were only "2x" and product people "5x," breaking traditional staffing ratios of engineers to product managers to designers.

To force deeper adoption rather than hiring around the gap, Ramp stopped hiring entirely for a quarter earlier this year despite growing "more than 100% per year." Glyman said the goal was to make AI fluency "your core competency" rather than "5% of your time."

Inspect: the internal coding agent reshaping engineering

Glyman said Ramp's coding agent, Inspect, is used in more than 85% of pull requests, with pull requests per engineer still growing roughly 40% per month. He described it as "merged with your team" rather than a separate tool, with access to code, data and its own virtual machines and browser.

"It's not just a great intellect, it's merged with your team." — Eric Glyman

He said customer support staff can now use Inspect to read a bug report, locate the code, write the fix, and email the customer back within minutes, work that previously required hiring additional engineers.

Glass, the "company brain," and radical transparency

Glass, Ramp's internal AI system, connects automatically to the company's data warehouses, analytics suites and support tools, giving a shared view of what is happening across the organization within individual access permissions. Glyman said 1,925 of Ramp's roughly 1,930 employees used Glass in the prior 24 hours. Shared "skill files" let successful workflows, such as a sales team's better research method, become defaults for others, and Glyman said this preserves institutional knowledge even after an employee leaves.

He gave the example of an expense-policy agent built with Glass that approves, declines, or flags incomplete items at what he said was 99% accuracy, learning and adjusting the policy itself as overrides occur. Glyman said this week Ramp began onboarding customers onto their own versions of Glass.

Why large firms die, and two extreme AI futures

Glyman argued the primary cause of death for large organizations is "indigestion," where coordination costs rise, information becomes siloed and hard calls go unmade. He said AI lets some large companies "move like lightweights again," pointing to Ramp's own growth accelerating as it scales.

Tan framed two extreme views on AI's effect on firm size: one where a single lab or firm captures nearly all future value, and another where hyperspecialization allows thousands of smaller firms to thrive. Glyman said he strongly favors the second view, based on recent data.

"I actually believe support worldview number two very strongly." — Eric Glyman

He also said that over the last two months, "Jevons paradox broke," meaning cheaper intelligence has not been offset by proportionally higher usage as it previously was, though he flagged this could be temporary.

Hiring philosophy and advice to his younger self

Glyman described using "Sutton's Law," named after a bank robber who said he robbed banks "because that's where the money is," as a hiring heuristic: look for people with demonstrated independent achievement, such as building a Minecraft business as a teenager or winning math competitions, rather than relying on resumes.

Asked what he would tell his younger self, Glyman said the most powerful thing a founder can say is "I need your help," describing a tendency among founders to avoid admitting problems and the value of structuring regular check-ins around growth, biggest problems, and planned next steps.

Written by AI from the video's transcript. It can compress, misattribute or miss context — the original video is the source. Not investment advice.

More from Y Combinator