JP Morgan

Windsurf built its own models for code bases too big for frontier models. Working with companies like Dell and JPMorgan Chase, whose single code bases exceed 100 million lines, Windsurf built models that read huge code bases in parallel across thousands of GPUs. They rank the most relevant snippets for each question.

A 100 million line code base is over 1.5 billion tokens, orders of magnitude beyond the largest context windows. Sending that much code to a frontier model is impossible, too costly and too slow.

Lenny's Podcast▶ 38:05

At JPMorgan's scale, not investing in engineering productivity costs more. JPMorgan spends $17 billion a year on software and has over 50,000 engineers. At that size, higher output per engineer makes not investing more costly.

Its code bases also exceed 100 million lines, which is the problem Windsurf's enterprise product was shaped around.

Lenny's Podcast

AI writing 90% of code does not mean 10X engineer productivity. Engineers also review, test, debug, design, deploy and navigate code. By Amdahl's Law, if coding is 30 of 100 units and drops to 3, the whole process only improves 27%. Windsurf sees over 30, maybe close to 40% productivity improvements.

Because the ROI of building technology rises, companies with high technology ceilings should hire more engineers. Quote: "If AI is writing over 90% of the code, that doesn't mean engineers are 10X as productive."

Lenny's Podcast▶ 1:04:20

Revolut targets JPMorgan's business and claims an edge on reach, price and interface. Nik Storonsky targets its business, including enterprise and investment banking. He says Revolut beats it on global reach, price and interface.

The business models differ. JPMorgan gives local accounts and points and leans on credit, while Revolut leans on subscriptions, transactions and commissions.

David Senra

Carnegie sold to J.P. Morgan for $480 million and later said he sold too cheap. Morgan paid $480 million for Carnegie's steel company and combined it with US Steel into the first billion dollar corporation. Carnegie later admitted to Morgan that he had sold out too cheap by $100 million, and Morgan replied, "very likely Andrew."

The $100 million figure comes from the House of Morgan story the host cites.

Founders▶ 50:34