Keep pulling the thread on Varun Mohan.
Exafunction's business model failed because the AI market consolidated around the Transformer architecture, eliminating the opportunity for a differentiated GPU virtualization provider.
Windsurf's pivot from GPU virtualization was triggered in mid-2022 when its primary customers, autonomous vehicle companies, began to fail to raise new funding rounds as the ZIRP era ended.
Windsurf's pivot to its Codium product was executed over a single weekend, abandoning a previous business that was generating a couple million dollars in revenue.
A non-developer employee at Windsurf used the company's own product to build internal applications, replacing over $500,000 in annual spending on external sales tools.
Windsurf has developed and deployed its own proprietary model, SWE1, which it claims is comparable in performance to frontier models for agentic workloads.
Windsurf's proprietary SWE1 model processes hundreds of billions of tokens of code daily.
For Windsurf's code autocomplete product, every 10-millisecond increase in latency reduces the user acceptance rate by one percentage point.
Varun, CEO of Windsurf, predicts that within 12 months, AI agents will become end-to-end assistants for the entire software development lifecycle, including design, deployment, and debugging.
Windsurf's predecessor company, Exafunction, was founded to build GPU virtualization technology based on the incorrect premise that hundreds of different model architectures would become popular.
Windsurf's current code review product was developed after several failed attempts in the previous year, including a Chrome extension and a parallel website beta.
Windsurf raised its Series A funding from Greenoaks based on its original GPU virtualization business model, prior to pivoting to CodeAI.
Windsurf's primary product strategy is to make existing engineers 10 times more productive, rather than enabling non-technical people to become engineers.