Uncapped with Jack AltmanVinod Khosla and Keith Rabois on Building and Investing in Enduring Companies | Ep. 40
CHAPTERS
Why Khosla Ventures changed Rabois’s investing: from zero AI to 70% AI
Keith Rabois opens with a reflection on rejoining Khosla Ventures and how it rapidly shifted his investing focus toward AI. He explains that without KV’s internal exposure and feedback loop, he’d have either missed the wave or invested recklessly.
- •Rabois had invested in 0 AI companies before rejoining KV; now ~70% of his investments are AI
- •KV partner meetings provided “learning by osmosis” through constant AI deal flow
- •He relied on internal “air cover” (Vinod, Sven, John Chu, Kanu) to evaluate teams and approaches
- •Iterative learning: investing → board work → understanding how AI companies actually get built
How Khosla and Rabois work together: first principles + brutal honesty
Jack Altman probes the day-to-day working relationship between Vinod Khosla and Keith Rabois. They describe a partnership rooted in first-principles reasoning, direct communication, and debate that clarifies assumptions rather than creating politics.
- •Their collaboration dates back to board work (Square) and earlier overlap (Slide)
- •Khosla prioritizes first-principles thinking as the basis for productive disagreement
- •“Brutal honesty” beats “hypocritical politeness” internally and externally
- •Shared style helps them partner effectively with highly ambitious founders
What KV actually spends time on: portfolio first, not firm admin or LP management
They break down the “pie chart” of partner time and explain KV’s operating philosophy. The firm minimizes internal management overhead and prioritizes active engagement with existing portfolio companies before chasing new deals.
- •Very little time spent on “operating the firm” (often <5%); mainly comp/hiring
- •Minimal time spent with LPs compared to many firms
- •Monday meetings intentionally start with supporting the current portfolio
- •KV frames itself as “venture assistance” rather than “venture capital”
Investor ethos then vs now: founder-friendly marketing vs doing what’s right for the company
Khosla critiques the industry trend toward performative “founder friendliness,” arguing it can harm companies by discouraging hard truths. They emphasize earned credibility—advice should come from people who have built and operated companies.
- •Key difference isn’t age; it’s whether an investor has “earned the right” to advise founders
- •Being “nice” can be counterproductive; strong founders want high-quality feedback
- •Founders should be able to disagree directly and still value counsel
- •They cite founder-driven ranking approaches (e.g., head-to-head/Elo-style evaluations) as more honest signals
Khosla Ventures vs Founders Fund: proactive company-building vs capital + optional help
Rabois contrasts KV’s hands-on model with Founders Fund’s more hands-off posture. Both aim to back contrarian, bold companies, but differ on how actively the firm intervenes to shape outcomes.
- •KV: proactive partner/“consigliere” helping build the company over 10–20 years
- •Founders Fund: provide capital, stay out of the way, help when asked
- •Analogy: even elite athletes have coaches—great founders can still benefit from active support
- •Shared north star: backing bold ideas whether fashionable or not
A practical framework for spotting great founders: extreme strengths, uncommon trait combos, and grit signals
Rabois lays out how he identifies exceptional founders early—often within minutes—either via top-0.1% strength in a single dimension or rare combinations of strengths. They discuss how grit and recruiting ability surface through stories and behavior rather than scripted interviews.
- •Rabois’s heuristic: “best I’ve ever met” on one dimension, or rare overlap of multiple strengths
- •Examples: Max Levchin’s tech + business combo; Jack Dorsey’s design + tech + strategy mix
- •Grit signals can come from personal stories demonstrating unusual drive under pressure
- •Recruiting is critical: the first 10–100 hires replicate the founder’s standards and culture
Khosla’s added lens: learning rate, critical thinking, and ethics as a hard constraint
Khosla expands the founder evaluation framework beyond “exceptionality,” focusing on learning velocity and the ability to reject bad ideas. He emphasizes ethics as non-negotiable and discusses methods to get beyond rehearsed interview answers.
- •Learning rate and open-mindedness: strong founders absorb new info but filter ideas rigorously
- •Khosla sometimes tests founders by taking positions he doesn’t believe, to probe thinking
- •Best evidence comes over time (e.g., YC partner view: what did they learn in last 3 months?)
- •Ethics is the one area founders can’t be deficient in; assessed mainly via references
Seeing alpha at seed: why consensus doesn’t reliably predict outcomes
They argue that seed-stage “hotness” isn’t a strong predictor of performance and that outliers often win. Examples like Rocket Lab and OpenAI illustrate how contrarian bets can look irrational at the time yet hinge on identifiable core insights (team density, capability).
- •Rabois: outcomes aren’t random, but many miss obvious founder/team signals
- •OpenAI thesis: unique concentration of research-grade talent outside Google/DeepMind
- •Khosla describes sending LPs an “apology letter” due to the investment’s outlier size/risk
- •At seed, consensus often tracks founder pedigree more than idea quality; outliers can outperform
AI investment themes: from copilots to “AI workers,” plus beyond-transformer bets
Khosla describes KV’s AI posture: prefer systems that do the work rather than assist humans, across many professions. They’re also investing in alternative approaches beyond transformers, while acknowledging it’s too early to know which techniques win.
- •Portfolio includes dozens of “AI worker” startups (oncologist, therapist, chip designer, engineer, etc.)
- •Strategic preference: avoid many “copilots”; humans become bottlenecks—aim for full task execution
- •They avoid investing in direct OpenAI competitors; emphasize loyalty to portfolio companies
- •Exploration of other paradigms: interpretability-led models, diffusion, neurosymbolic ideas, category theory, real-world/physical models
Hard problems in applied AI: intuition, world models, and eliminating hallucinations in high-stakes domains
They discuss where current models fail and what architectures might be required to win in sensitive applications. Khosla highlights “intuition” and physical-world understanding as pivotal, and points to hallucination-free design as essential in regulated or safety-critical settings.
- •World models/embodiment: high conviction they will work; competition still “up for grabs”
- •Example: General Intuition trained on gaming data to replicate tactical “intuition” in video prediction
- •Hallucinations: some domains can tolerate them, but finance/insurance/banking cannot
- •Winning agentic products may require architectures optimized for knowing when not to hallucinate
How AI changes company-building: growth expectations, PM roadmaps, talent comp, and GTM playbooks
Rabois argues AI companies are built differently due to unprecedented growth rates and rapidly shifting capabilities. They discuss how classic roles and processes (like traditional PM roadmaps) break down, and why talent economics and sales motions must be rethought.
- •AI-era growth resets expectations (e.g., 0→$50M enterprise revenue much faster than historic norms)
- •Traditional PM function and 12-month roadmaps don’t fit monthly capability shifts
- •Customer acquisition paired tightly with research (OpenAI-style) as a new organizational pattern
- •Talent comp is structurally different; startups must redesign incentives, costs, and recruiting strategy
- •GTM shifts: top-down CEO sales can work unusually well due to “AI pressure” from boards/executives
Beyond AI: fintech repeatability, energy/sustainability, manufacturing onshoring, and defense tech momentum
They broaden the lens to KV’s other enduring themes. The conversation covers KV’s historical strength in financial services, optimism in energy and sustainability, AI-driven manufacturing transformation, and the expanding defense-tech opportunity shaped by geopolitics.
- •Fintech: claim that each KV fund has been returned by at least one financial services investment (e.g., Square, Stripe, Affirm, Upstart, Ramp, Aven)
- •AI in finance is constrained by risk/hallucination concerns, but companies like Aven show speed gains (e.g., HELOC-to-card workflows)
- •Manufacturing: AI can reduce labor cost by automating engineering/system design, enabling onshoring beyond just robotics
- •Defense: early bets before it was “cool” (Rocket Lab, Hermeus, Mach, Varta); view tech as critical to national security competitiveness
Politics and posting on X: principles, influence, and the China/AI regulatory tension
Jack asks why they engage politically online and how their views evolved. Khosla frames his independence and anti-Trump stance around values, while both emphasize avoiding “convenient” belief shifts, and they converge on concerns about AI regulation and competition with China.
- •Rabois: uses X to challenge bad ideas; wishes he had more time for rigorous evidence-building
- •Khosla: spends limited time, engages when principles are violated; cites criticism of price controls/interest-rate caps
- •Khosla’s political evolution: Republican → Independent, driven largely by climate and principles
- •Shared concern: excessive or fragmented (state-level) AI regulation could weaken the US vs China
- •They describe a “techno-economic battle” framing and argue the US must maintain AI leadership
