Ramp's AI build factory: 75% of pull requests now written by agents

Key points

Key takeaways from Lenny's Podcast summit talk with Ramp CPO Geoff Charles (September 2026):

Speed is a systems problem. Charles argues product velocity is about removing bottlenecks in the build process, not coding faster: Formula 1 pit stops fell from 67 seconds in the 1950s to 1.8 seconds today, and with 90% of an F1 car's parts replaced yearly, product teams face the same relentless bar — the best driver can only perform up to the level of the system.

The bottleneck has shifted to product. With coding largely automated, engineering is no longer the constraint — defining, collaborating, coordinating, testing and releasing now are, so Charles urges PMs to 'be as lazy as engineers' and invest in the software factory around them.

Customer insight agents. An early 'hate channel' posting real negative customer quotes got out of hand, so Ramp rebuilt itas a customer insight agent combining ETL pipelines, vector search and clustering over Gong, Zendesk and LogRocket data — accessible org-wide, even as a daily 'hate podcast'.

AI-native product definition. Internal agent Glass connects to Snowflake, user research and Ramp's codebase, turns concepts into evidence-backed specs and working prototypes grounded in the design system — letting AI absorb the PM's 'is this possible?' questions that used to go to engineers.

Anyone at Ramp can now code. Internal coding agent Inspect runs in Slack, spins up provisioned, deployed previews in under 5 seconds,and has run a million sessions; 75% of Ramp's pull requests are built by Inspect, including a thousand last month from non-engineers.

Automated review and testing. Review Buddy understands the codebase, quality checks and security concerns, routes the right human reviewer in,and auto-handles 93% of pull requests; browser-based QA agent Testo spins up products in 100 combinations from production data and caught 425 bugs in the last 30 days.

Coordination via 'gadgets'. With 'every question an API', agents read Notion, Slack and Linear to answer status and sales questions with evidence, update roadmaps, ping late owners and draft help-centre articles and launch emails — 85% of questions directed at PMs are now fully answered by AI,and 60% of identified UX issues are fixed within 24 hours.

Three futures for product leaders. Charles sees PMs evolving along three tracks: technical PMs who build the internal factory, tastemakers holding the bar for great product,and GMs owning business outcomes across product, marketing, sales, growth and operations.

Read more: Lenny's Podcast (YouTube)