How Windsurf grew to $1.3B by pivoting fast and betting on future tech

Executive overview

Most startups die by doing the wrong thing for too long. Varun Mohan built Windsurf by treating failure as signal and pivoting the company overnight — twice — to chase where technology was heading, not where it stood.

The core practice: set visionary goals, but validate through tractable intermediate steps. Build lean, stay paranoid, and never confuse current product-market fit with a durable moat.

If everything is working, you're not taking enough bets.

From GPU virtualisation to AI code editor

• Founded in 2021 as ExaFunction, building GPU virtualisation and compiler tech
• Reached ~$2M revenue with 8 employees, but growth felt ad hoc and the ceiling was low
• GPT-3.5 in mid-2022 convinced them generative models would commoditise infrastructure complexity
• Decision to pivot made over a weekend; team started on the new direction Monday morning
• Built Codium — a free VS Code extension with their own model — as the tractable first step
• MVP shipped in under two months; hundreds of thousands of users followed quickly
• Pivoted again mid-2024 to build a full IDE (Windsurf) to control the agentic experience end-to-end
• Regret: not doing each pivot three months earlier

Why pivot fast instead of iterating

• Incremental 20-30% improvements on a low-ceiling idea don't compound into a big company
• Scrapping is better than clawing; every pivot is a chance to 10x the company's size
• Having raised $28M made the decision harder — but the logic was the same
• Writing the company off as zero removes fear; any outcome above zero is upside
• Intellectual honesty with the team matters: good people don't leave when you chart a credible new path

Building for where technology is going

• Consumer GPU compute grew from 10 to 700 teraflops between 2017 and 2022 — models were always going to get better
• Short-term model-specific heuristics have massive diminishing returns; the next model makes them irrelevant
• Autonomous vehicles taught the lesson: bet on where the technology will be, not where it is today
• Windsurf's goal: reduce the time to build technology by 99% — deliberately ambitious, not achievable near-term
• Visionary goal + tractable intermediate steps is a feature, not a compromise

Running lean and scaling with urgency

• Company runs as the smallest it could be for its level of ambition
• Everyone should have too much to do; being underwater forces rapid prioritisation
• Hire only when the team is genuinely underwater; stop scaling the moment they surface
• Varun still personally interviews every hire to preserve culture at ~200 people
• Excess headcount creates competing priorities — people find things to fill time, which fragments focus

Product-market fit and staying paranoid

• True PMF is when inbound demand exceeds what any of you can handle — it's obvious, not a metric
• The GPU virtualisation business had revenue but no PMF: no clear path to 10x or 100x scale
• PMF is not durable: without continued innovation, today's differentiation becomes tomorrow's commodity
• Telling the company "we're probably going to fail" is a feature — it sustains the urgency to find the next thing
• Listen to customers about pain, but not about how to build; transformational solutions are often unexpected
• Windsurf teams build on Windsurf — internal use is the fastest feedback loop