newsfilter
← Interviews
Y Combinator54 min

How To Build The Future: Eric Glyman, Co-CEO of Ramp

Eric Glyman, Garry Tan, Gary

Fireside Chat, Interview
  • Ramp is transitioning from a "system of record" to an "AI harness" via its "Glass" platform, which is currently being onboarded to customers "this week" to enable agentically interacting with corporate data; the system will connect to core infrastructure like data warehouses and Zendesk to store "skill files" for institutional memory and allow semantic ROI measurement linking specific spend to outcomes.
  • Strategic plans include a growth acceleration where the company's expansion rate is "more than 25% faster now than a year ago" and a hiring pause for "a whole quarter" to prioritize AI literacy as a "core competency," with internal usage of the "Glass" system projected to reach "1,925" out of "1,930" employees within "24 hours."
  • The "Inspect" coding agent is projected to handle "85% plus" of all pull requests, which are expected to grow at "40% every month," while non-engineer pull requests rose from "functionally zero" two years ago to "7,000" in Q2 and are forecast to rise further over the next "two years" toward a "100x engineer" productivity standard that reduces development time from "three months" to "minutes."
  • Agent capabilities will utilize "super vision" and full code access to locate bugs, write changes, and respond to customers within "15 minutes" without human intervention, often accepting "10% longer" process times to achieve a "50% cut in price" by deploying simpler models rather than frontier models.
  • The policy agent is expected to operate with "99% accuracy" on expense approvals and flagging, while "MCP usage" is forecast to "explode" at a growth rate of "10% per week" to become a dominant platform interaction method in the "not too distant future."
  • Customer economic benefits include saving "7.5% per year" on expenses, which is a multiple of interchange returns, and the ability to reduce waste by "50 fewer dollars" per transaction, with future model shifts enabling most "frontier model" spend to be replaced by "simpler models" via "smarter defaults."
  • Market adoption predictions suggest the "top 1% of spenders" currently account for "80% of the spend on AI tokens," but the remaining "99%" are expected to adopt within "nine to 24 months" as the "Jevons Paradox" breaks, potentially leading to "thousands or tens of thousands" of hyper-specialized firms rather than a single monopoly.
  • Future industry dynamics anticipate "intelligence" becoming "almost free" to enable previously impossible businesses, with large companies expected to "defy gravity" and grow like "lightweights" by reducing coordination costs and siloed information, although risks include internal politics and "not invented here" sentiments that must be overcome for the "99%" to succeed.