EO StudioHow I Built a $1.3B Startup by Pivoting Fast | Windsurf, Varun Mohan, Co-Founder & CEO
CHAPTERS
- 0:00 – 1:00
Fail fast as a startup superpower: obvious feedback and rapid pivots
Varun frames startups as repeated failure cycles—and argues that failing is not just inevitable but necessary when you’re pushing the frontier. Clear failure is “freeing” because it removes ambiguity and forces decisive change. He also reflects on wishing he had pivoted even earlier, emphasizing humility and validation over attachment to ideas.
- •Failure provides clearer signals than slow, ambiguous progress
- •If nothing is failing, you’re likely not taking enough bets
- •Validate ideas; don’t confuse idealism with correctness
- •Pivoting is positive—staying on the wrong path is worse
- •Regret: not pivoting a few months earlier
- 1:00 – 3:01
What Windsurf is today: AI-powered IDE + the company’s origin story
He introduces Windsurf as an agentic AI IDE used by over a million developers, with rapidly growing monthly active users. He traces his path from MIT to Nuro, then to founding the company in 2021—initially far from code AI.
- •Windsurf: agentic AI IDE for developers and non-developers
- •Scale: 1M+ users and strong MAU growth
- •Founder background: MIT (2017) → Nuro (autonomous vehicles)
- •Company founded in 2021; early focus wasn’t code AI
- •Early product: GPU virtualization/compiler tech
- 3:01 – 4:01
Choosing smaller, high-agency teams over big tech prestige
Varun explains that his career choices were driven less by brand and more by impact and mission. He progressively interned at smaller companies to maximize meaningful contribution and be close to visionary work.
- •Motivation: meaningful role on a visionary team
- •Intern progression: LinkedIn → Quora → Databricks (early)
- •Didn’t optimize for big-company security or status
- •Chose Nuro to work on the future of robotics/AVs
- •Core driver: motivated people + frontier technology
- 4:01 – 5:02
Lessons from autonomous vehicles: build for where compute and models are headed
He shares a key insight from AV work: capabilities change exponentially as compute scales. Betting only on what works today makes products quickly obsolete; you need to design for where the tech will be in a few years.
- •Compute increased dramatically (consumer GPU teraflops)
- •Models improve fast; product assumptions must anticipate that
- •Don’t build only for today’s constraints/capabilities
- •AVs became increasingly ML-driven over time
- •Strategic lesson: future-proof bets beat short-term hacks
- 5:02 – 5:33
ExaFunction era: GPU virtualization success—and why it still wasn’t the right business
The company’s original mission (ExaFunction) aimed to virtualize GPU computation and achieved a couple million in revenue with a small team. But Varun explains why the business felt ad hoc and capped, especially as generative AI threatened to commoditize the infrastructure layer they were building.
- •ExaFunction: virtualizing GPU computations (exafunction concept)
- •Reached a couple million revenue with ~8 employees
- •Growth felt unscalable and overly ad hoc
- •Generative models suggested infra complexity would commoditize
- •Couldn’t see a path to 10–100x growth toward billions
- 5:33 – 7:33
Killing a $28M-funded direction: the weekend decision to start from scratch
Varun recounts the difficult decision to scrap the prior business despite funding and traction. He and his co-founder decided over a weekend, told the team Monday, and fully committed—because startups can rarely do two important things at once.
- •Pivot required abandoning sunk cost: $28M raised + existing revenue
- •Belief: pivots can create a chance to 10x company potential
- •Incremental improvement on a low-ceiling idea doesn’t matter
- •“Rip the Band-Aid”: one clear direction, not two strategies
- •Team alignment through honesty even if some attrition risk
- 7:33 – 9:50
Startup moat redefined: right bets + great people + time
He challenges the idea that early startups have durable moats; with only a handful of people, any technical lead is shallow. The real moat is consistently working on the right thing with a high-quality team long enough—supported by intellectual honesty and transparent leadership.
- •Early “moats” are shallow; few engineering-years exist
- •Real moat: correct strategic bets + strong team execution over time
- •Transparency about strategy changes keeps culture strong
- •Good people stay when there’s a credible path forward
- •Avoid doing things just to fit investor/industry narratives
- 9:50 – 11:51
From Copilot inspiration to Codium: the 99% faster software ambition and MVP discipline
Seeing Copilot as the “tip of the iceberg,” the team set an audacious goal: reduce software-building time by 99%. They paired that vision with tractable intermediate steps—shipping a free IDE extension powered by their own models, built and released in under two months.
- •Copilot revealed a larger opportunity beyond autocomplete
- •Vision: move from suggestions to generating substantial code/PRs
- •Set ambitious target: 99% reduction in build time
- •Used infra strengths to train/run models and ship free extension
- •MVP shipped in <2 months with clear intermediate milestones
- 11:51 – 13:21
Zero to pull: rapid adoption, enterprise inbound, and what ‘product-market fit’ really means
Inbound demand from companies quickly outpaced what the team could handle, signaling real pull and the need for a go-to-market function. Varun also critiques “product-market fit” as a static concept—arguing it can disappear quickly without continued innovation and paranoia.
- •Enterprise inbound became unmanageable—clear market pull
- •Hired GTM once demand proved real
- •Security and deployability mattered for large orgs
- •PMF isn’t permanent; competition can commoditize advantages
- •Operating mindset: paranoia + urgency to stay differentiated
- 13:21 – 13:51
Scaling to enterprise needs: secure deployment and massive codebase personalization
As large customers arrived, the product had to handle complex, gigantic codebases and deliver more personalized suggestions. The company rapidly scaled to ~100 customers within months by building capabilities that fit real enterprise constraints.
- •Customers with tens of millions of lines of code
- •Need: personalized suggestions across complex repositories
- •Security-driven demand: ability to run models securely
- •Fast customer growth: 0 → ~100 customers in months
- •Product expanded to handle enterprise-grade complexity
- 13:51 – 14:52
Becoming a product company: UX details, latency, and avoiding “customer-driven mediocrity”
Varun explains the shift from infrastructure thinking to product craftsmanship, where small UX details like latency and machine impact matter. He also warns that blindly building exactly what customers request can lead to incremental, worse products—customers define problems, not necessarily solutions.
- •Product details (latency, responsiveness) became critical
- •On-device computation tradeoffs can harm UX
- •Transition: infra-first mindset → product-first mindset
- •Customer obsession can backfire if it prevents paradigm shifts
- •Listen to customers on pain; don’t outsource design strategy
- 14:52 – 15:52
Dogfooding and tight feedback loops: building Windsurf using Windsurf
He describes how the team stays close to user pain through direct channels and constant internal use. Because employees build the product using the product, internal feedback is immediate—and often predictive of external user satisfaction.
- •Frequent review of user feedback via online channels
- •Company-wide dogfooding: build Windsurf on Windsurf
- •Internal friction is treated as a leading indicator
- •Fast iteration enabled by constant real usage
- •Feedback loops are both external and deeply internal
- 15:52 – 18:23
Running ~200 people like a small team: lean execution, prioritization, and culture filters
Varun outlines an operating philosophy of staying as small as possible for a given ambition: everyone should be “underwater,” forcing prioritization and making hiring a deliberate response to real load. He also maintains culture by personally interviewing every hire, and stresses focus on doing one thing extremely well despite scaling challenges.
- •Lean-by-design: smallest team possible for the ambition
- •“Underwater” as a forcing function for prioritization and hiring
- •Process added to speed cross-functional shipping (not slow it)
- •Risk with scale: too many priorities and make-work
- •CEO interviews all hires to preserve culture and standards
- 18:23 – 21:37
Optimism vs realism: existential urgency, idea killing, and building for the tech curve
He closes with the tension every startup must manage: irrational optimism to compete with giants, paired with uncompromising realism to kill bad ideas fast. He argues teams should build for where technology is going, avoid brittle short-term heuristics, and instead invest in compounding advantages from user-driven learning and better experiences.
- •Operate with existential urgency without paralysis
- •Balance: irrational optimism + uncompromising realism
- •Kill ideas quickly; most ideas are bad
- •Build for the future tech curve, not today’s constraints
- •Avoid short-term model “hacks”; focus on user-learned experiences
