The Twenty Minute VCNabeel Hyatt, GP @ Spark Capital: To Win in AI, Investors Need to Change Their Approach | E1255
CHAPTERS
- 0:00 – 0:50
VC industrialization, career incentives, and how too much capital breaks companies
Nabeel opens with a critique of modern venture as an industrialized machine driven by junior incentives rather than long-term outcomes. He argues that many actors optimize for promotions and short-term markups, and that oversized rounds can actively harm young companies.
- •Industry increasingly run by principals/associates optimizing for promotion, not exits
- •Startup creation becomes weekly “arbitrage” to graduate incubators and raise seed
- •Short-term markup chasing distorts decision-making
- •Too much capital can change behavior and “mess up” a company
- 0:50 – 3:41
AI forces a mindset shift: from SaaS “puzzles” to AI “mysteries”
Harry and Nabeel frame AI investing as fundamentally less knowable than the prior SaaS era. The old playbooks and metrics worked when venture could be systematized, but AI’s rapid model shifts make outcomes harder to precompute.
- •Founder–market fit has an analogue: VC–market fit
- •B2B SaaS era rewarded metric-driven optimization and repeatable playbooks
- •AI landscape changes too fast to rely on “spreadsheet certainty”
- •Investing becomes a ‘mystery’ requiring judgment under fog-of-war
- 3:41 – 6:37
Rebuilding venture firms for rapid creativity (and why most won’t adapt)
Nabeel argues firms must reorient around small, founder-like teams making subjective, craft-based bets. He explains why LP expectations and internal power dynamics slow adaptation even when the market demands it.
- •AI era requires ‘rampant creativity’ and nuance, not checkbox investing
- •LPs prefer stability even when the environment is unstable
- •Firm evolution is slow because fund cycles and mandates lag reality
- •Best support comes from being a curious, product-using sounding board
- 6:37 – 8:58
Why old heuristics break: revenue speed, dashboards, and the partnership politics behind them
They discuss how classic quality signals (ARR milestones, time-to-revenue) are less predictive when AI products can spike fast and fade faster. Nabeel claims these heuristics often exist to win internal partnership approval, not to predict durable value.
- •Fast ARR in AI can be real—and still be ephemeral
- •Dashboards/heuristics often serve internal consensus-building
- •Large partnerships push toward standardization and political selling
- •Pre-exit metrics can become ‘false prophets’ that fuel politics
- 8:58 – 12:53
The incentive trap: markups, “packaging,” and the ‘Brita filter’ inbound funnel
Nabeel connects the principal/associate ladder to markup-chasing behavior: invest one step ahead of the next-stage investor’s taste. He rejects venture as a packaging industry and criticizes high-inbound, fast-no funnel investing as transactional and shallow.
- •Promotion cycles reward quick markups over long-term alignment
- •‘Packaging’ is a losing strategy when the future shifts quickly
- •‘Brita filter’ investing: maximize inbound, filter fast, stay transactional
- •Pattern matching breaks down when markets are discontinuous
- 12:53 – 18:07
Is venture commoditized? Competing on deployment vs performance, and living with expensive deals
They debate whether venture has become a low-margin commodity business, and how that depends on market conditions and competitor quality. They also discuss founders taking extreme terms (high prices, common stock) and how conviction interacts with valuation.
- •In fast-change eras, only a subset of firms truly compete on the same deals
- •Hard to win when others optimize for deployment and price-insensitivity
- •Best deals can still be highly priced—rules have many exceptions
- •Valuation often tests conviction, but there’s always a breaking point
- 18:07 – 19:29
Coaching inside partnerships: self-actualization, truth-seeking, and “brother/sister’s keeper”
Nabeel describes mentorship as helping partners find their own investing superpowers rather than copying senior partners. Spark’s internal debate aims to be a search for truth, with partners calling out each other’s blind spots and recurring failure modes.
- •Mentorship is about discovering strengths, not producing “mini-me” investors
- •Partners should know your weaknesses better than you do
- •Healthy debate isn’t politics; it’s collective truth-seeking
- •Long-term performance requires honest feedback loops
- 19:29 – 22:19
When investing felt broken: COVID-era Zoom venture and why early vs growth are different sports
Nabeel shares that COVID-era speed and remote decision-making made him question the industry, so he wrote very few checks. He explains why growth investing can support more hierarchy and diligence, while early-stage requires different instincts and process.
- •Zoom-era ‘write in hours’ dynamics undermined relationship-based judgment
- •Nabeel slowed down dramatically rather than forcing bad decisions
- •Growth can rely more on diligence, numbers, and customer calls
- •Early-stage lacks data, so craft and founder-read matter more
- 22:19 – 25:23
VC as service vs ‘founders don’t need help’: raising the bar beyond mediocrity
Responding to Keith Rabois’s view, Nabeel argues founders should want investors who are deeply engaged—unless they can’t find them, in which case ‘no-op’ is acceptable. He emphasizes that high service is incompatible with high volume, making the model a deliberate choice.
- •‘Fine’ board members would be unacceptable in most other life contexts
- •Great investors do detailed, context-specific work (not recycled advice)
- •If you can’t get great help, choose a VC who does no harm
- •High service requires low deal volume; it’s a strategy trade-off
- 25:23 – 29:41
What happens when Nabeel loses faith in a founder: conflict avoidance and taste vs speed
Nabeel explains two patterns that lead to disappointment: founders who avoid conflict and founders miscast on the spectrum between execution speed and taste/judgment. He argues the right founder profile depends on whether the company is executing in a crowded arena or creating a new market.
- •Conflict avoidance is hard to detect pre-investment but deadly in execution
- •Founders must balance speed with taste—both are required in AI
- •Over-indexing on speed can become shiny-object chasing
- •Market type (execution vs creation) dictates the founder “casting” needed
- 29:41 – 36:19
Durability in a world that resets overnight: continuous reinvention, market-size skepticism, and product as a window into people
They explore how value can persist when platforms and models can wipe out moats quickly. Nabeel downplays traditional TAM analysis, looks for new behaviors that “sear into your brain,” and reframes product-focus as a way to understand the builders’ judgment rather than fetishize UI.
- •Sustainability comes from founders who continually reinvent, not static moats
- •Market size is rarely decisive; behavior change and 10x experience matter more
- •Product evaluation reveals how founders think and decide
- •Avoid judging early-stage via generic templates; ask role-appropriate questions
- 36:19 – 42:41
Deal flow reality: meeting volume, chemistry-first calls, in-person conviction, and the Figma miss
Nabeel outlines how many companies he touches weekly, why 30-minute first calls beat 1-hour calls, and why in-person matters for his process. They discuss price sensitivity as a conviction test, the dangers of oversized rounds, and his biggest price-related regret: passing on Figma.
- •High inbound exists, but real evaluation requires deeper time and chemistry
- •Half-hour first call to assess fit; then go deep (walks, long sessions)
- •Rarely invests without meeting in person (unless deep prior relationship)
- •Too much capital can alter a company’s trajectory; check size matters as much as price
- •Biggest regret due to price/check dynamics: Figma (and the missed journey)
- 42:41 – 43:59
Secondaries and liquidity: why focus beats financial engineering
Harry presses on secondaries and liquidity as IPO timelines stretch. Nabeel acknowledges secondaries can happen but argues the best use of time is still primary underwriting—building curiosity, trying products, and staying open to founders’ non-consensus visions.
- •Secondary sales aren’t ‘never,’ but they’re not the core craft
- •Time is the limiting resource; focus on doing the main job exceptionally well
- •Curiosity and product engagement are essential to seeing the future early
- •Avoid stacking extra roles (growth, conferences, constant market ops) at the expense of underwriting
- 43:59 – 50:54
Three categories of AI startups: adaptation, evolution, revolution (and where Spark wants to play)
Nabeel introduces a lens borrowed from the mobile era: adaptation (AI coat of paint), evolution (new workflows/behaviors), and revolution (new platforms made possible by the tech). He argues the market is flooded with shallow adaptation/evolution plays, while Spark prefers evolution and especially revolution despite higher risk.
- •Adaptation: incumbents and copycats bolt AI onto existing products
- •Evolution: AI-native workflows that change user behavior (e.g., Granola, Descript)
- •Revolution: platform shifts that only exist because of the new tech (mobile-era Uber analogue)
- •Most startup supply is shallow arbitrage; true disruptive bets are rarer
- •Spark avoids adaptation, leans toward evolution/revolution for meaningful outcomes
- 50:54 – 1:15:17
Models, interfaces, and the agent future: Anthropic’s edge, DeepSeek, data exhaust, and choosing ‘hard jobs’
Nabeel defends investing in both foundation models and applications, arguing sustainable advantage comes from full-stack iteration, customer interface ownership, and learning from usage (“data exhaust”). He downplays DeepSeek as a conceptual surprise, then extends the discussion to agents: many are near-term arbitrage unless they target genuinely hard problems with multiple acts of innovation; the episode closes with reflections on geography (SF vs Europe), raising-path conversations, and a quick-fire on mindset and uncertainty.
- •Bull case for Anthropic/OpenAI includes UI + scale-driven feedback loops, not just model quality
- •DeepSeek reinforces that capital alone isn’t the moat; execution + taste still matter
- •‘Data exhaust’ and expert workflows (e.g., Descript edits) can become defensible learning advantages
- •Agents can zero out markets via commoditization unless the startup has deeper second/third acts
- •Advice to founders: pick problems not solved by today’s models; build for a decade, not an incubator demo
- •Geography: SF concentrates AI talent and information flow; Europe faces coordination and incentive challenges
- •Raising-path coaching: aim to exceed expectations with a compelling story, not just hit obvious metrics
- •Quick-fire themes: don’t evaluate companies by underlying models alone; navigate uncertainty without terror
