Uncapped with Jack AltmanHow AI Is Rewriting Seed Stage Investing with Kevin Hartz & Bennett Siegel | Ep. 49
CHAPTERS
- 0:00 – 0:26
AI is killing “traditional software” pitches: systems of intelligence and action
Bennett and Jack open by noting that nearly every seed pitch now claims to be an AI company, and truly “traditional software” decks are rare. They discuss why code generation compresses the value of old workflows, while creating new budget categories for AI-native applications.
- •Everyone adding “AI” to decks vs real AI-native businesses
- •Codegen/vibe-coding reduces the moat of workflow apps built by large engineering teams
- •Shift from systems of record toward systems of intelligence and systems of action
- •Application-layer opportunities still exist, but require real differentiation
- 0:26 – 1:16
How A* Capital formed and what it’s optimizing for at seed
Kevin and Bennett explain how they first connected through Ramp and how A* was started to be a founder-partner at the earliest stages. They outline the firm’s evolution over five years, including closing a third fund and reaching ~$1B AUM.
- •Kevin met Bennett via Ramp at the seed stage (pre-A*)
- •A* founded with a third partner, Gotham, to build a new kind of early-stage VC firm
- •Focus on close partnership with founders from the earliest days through growth
- •Five-year progress: third fund closed, ~1B AUM, differentiated early-stage posture
- 1:16 – 5:27
Why large funds behave differently: fees, incentives, and seed as “option value”
The conversation turns to why multi-billion-dollar funds have moved into seed and how incentives shift as fund size grows. Kevin and Bennett argue that traditional “2 and 20” economics distort behavior at scale and encourage seed investing as a portfolio of options rather than deep partnership.
- •If fees were capped, big-fund behavior would change dramatically
- •Mega growth funds increasingly resemble indexing; question whether they deserve 2 and 20
- •Big funds come to seed for option value: build a basket, then double/triple down on winners
- •Founders may dislike being treated as an “option,” even if valuations are attractive
- 5:27 – 8:10
Competing with frothy pricing: what seed specialists can still sell
Jack poses the core challenge: founders may prefer higher valuations and larger checks even if partnership quality is lower. Kevin and Bennett explain how boutique seed firms win by offering competitive dilution plus real support, and they reflect on how frothy markets make “help in hard times” hard to communicate.
- •Founder counterpoint: “10 at 100” can beat “3 at 30” due to more runway
- •Boutique agility: A* can’t win every deal, but can win with founders seeking true partners
- •Risk of too much capital too early: more rookie mistakes and less oversight
- •2021 as precedent: high-price, low-involvement capital often produced poor outcomes
- 8:10 – 11:12
Valuation inflation and the ‘mother of all bubbles’ framing
Kevin predicts the market is heading toward an enormous AI-driven bubble, comparable to past platform shifts (PC, internet, mobile). Bennett adds that despite likely “carnage,” durable winners will emerge, while escalating seed/A/B pricing raises the bar for companies to grow into their capital structures.
- •AI as a platform shift driving a historic boom—and eventually a painful bust
- •Past bubbles still produced generational winners (Amazon/Google after 2000)
- •Bubble signals: faster rounds, speculative behaviors, SPVs—yet real revenue growth exists
- •Step-function inflation: seed 20–30 post → 40–50; Series A ~100 post → ~250; Bs often $500M–$1B
- 11:12 – 13:09
Founders are getting younger—and AI resets the advantage curve
They describe a visible trend toward younger founders, drawing parallels to Jobs, Gates, Zuckerberg, and the Thiel Fellowship’s thesis. Bennett argues AI is rewriting go-to-market and product rules for everyone, making early adopters (often younger builders) especially well positioned.
- •Clear trend: founder ages skewing younger, including teen founders
- •Historical parallels: Jobs/Gates at 19; acceleration through Zuck era and Thiel Fellows
- •AI makes prior SaaS playbooks less relevant; everyone is learning in real time
- •Young founders often become first adopters and fastest iterators in new paradigms
- 13:09 – 16:31
Mapping talent, not markets: how A* sources founders and spots “nodes”
Bennett explains A*’s talent-first approach: find dense networks where exceptional people cluster and where founder-quality repeats. They discuss signals like elite competitions, top universities, and founder factories such as Palantir, plus the importance of non-consensus mindsets in certain alumni groups.
- •Seed requires founder-centric selection; markets are often unknowable early
- •Labs now compete at the application layer; need to identify white space/runway
- •Tactical sourcing: talent-dense nodes (universities, accelerators, elite companies)
- •Founder factories and patterns: Palantir highlighted for producing many operator-founders
- •Non-consensus mindset as a founder trait (e.g., early Palantir/Anduril employees)
- 16:31 – 17:43
The rise of AI researcher founders—and how to evaluate them
They note a new class of founder: researchers leaving labs or PhD programs and raising unusually large rounds. While historically seen as an anti-pattern (high intellect, low commercial orientation), today’s market has counterexamples; Kevin emphasizes careful listening for clarity, intent, and precision in communication.
- •Researchers as founders are newly common and can raise massive capital
- •Old anti-pattern: academics often lack commercialization instincts
- •New reality: AI breakthroughs enable researcher-led companies to scale faster
- •Evaluation cues: precision of language, preparation, and ability to articulate the business clearly
- 17:43 – 19:59
Relationship-driven seed vs “shotgun marriage” rounds: what performs better
They compare investing with prior founder relationships versus meeting founders during a fast-moving fundraising process. Using Decagon as an example, they argue history and context can make it easier to underwrite even when the idea is nascent; proprietary-ish deals have driven disproportionate returns for A*.
- •Decagon example: strong team + white space; idea initially not fully formed
- •Prior relationships improve underwriting and conviction in ambiguous seed moments
- •In-round investing can be a “shotgun marriage” requiring speed and less information
- •A* says >50% of seeds are pseudo-proprietary; those have performed better on average
- 19:59 – 22:54
Why seed investing is uniquely hard—and why seed firms don’t persist
Kevin and Bennett unpack why seed is both high-effort and difficult to generate top fund returns: most companies won’t be large enough to matter. They also describe why seed firms often disappear: failure to refresh networks/strategies, or ‘graduating’ into later-stage funds once successful.
- •Seed difficulty: minimal product/roadmap; investing mostly in people amid high uncertainty
- •Effort intensity: earning the right to invest, then heavy post-check support
- •Power law reality: sub-$1B outcomes often don’t move the fund needle
- •Seed firm attrition: lack of adaptation/network refresh, or upward drift into opportunity/growth funds
- 22:54 – 27:31
Concentration and follow-ons: pro rata isn’t enough
They argue that returns depend heavily on follow-on decisions and concentrating capital into the few breakouts. A* describes a reserve-heavy model where more than half of fund dollars go in after the seed, and they discuss the challenge of identifying ‘what great looks like’ across stages.
- •Reserve-heavy strategy: >50% of A*’s fund deployed after seed
- •“Peanut butter pro rata” across the portfolio can dilute returns potential
- •Follow-on ‘fault line’ decisions can matter more than the initial seed check
- •Need stage-specific pattern recognition: understanding greatness from seed through growth
- 27:31 – 30:02
AI rollups: why they sound easy and usually aren’t
Jack asks about the trend of buying existing businesses and ‘adding AI’ to improve margins. Bennett is skeptical: rollups are culturally and operationally hard, venture math can be misaligned (owning too little of an asset), and few models have shown venture-like returns despite creative structures.
- •Rollups underestimate integration difficulty: culture, processes, and change management
- •Founder-friendly vs VC-friendly: founders may keep large upside while VCs own limited %
- •Venture investors focus on top-line growth; rollups often require bottom-line precision
- •Complex holdco/opco structures exist, but venture-scale monetization remains unproven
- 30:02 – 31:15
Where AI rollups might work—and why A* won’t do them
Bennett identifies narrower rollup categories that could benefit from AI-driven labor automation (e.g., ticket-heavy or services workflows), while warning that most value comes from buying at the right multiple—an area where VCs are historically weak. They mention Bending Spoons as an example of a strong operator, but reaffirm A* won’t pursue rollups.
- •Potentially interesting rollup targets: services-like businesses with labor automation potential
- •Examples mentioned: accounting, HOA management, ITSM/ticket deflection
- •Core value often comes from purchase multiple discipline, not technology alone
- •A* explicitly: no rollups; Bending Spoons cited as an exceptional integrator/operator
- 31:15 – 33:01
AI vs traditional software: labs as competitors and the new application battleground
They revisit the idea that labs increasingly compete at the application layer and that ‘traditional software’ is under pressure, reflected in public market drawdowns. Bennett frames the opportunity as building systems of intelligence/action; Kevin emphasizes continuing to back independent application companies where teams can build real barriers.
- •Labs are no longer pure infrastructure; they compete in applications too
- •Traditional workflow apps are vulnerable as building software gets cheaper/faster
- •AI-native apps target new spend categories (intelligence + action), not just workflows
- •Incumbents can win if they re-architect quickly (Ramp cited as a strong adapter)
- 33:01 – 35:38
Robotics and physical-world AI: excitement, defensibility, and valuation risk
They discuss why hardware and robotics are increasingly attractive, partly because real-world deployment creates defensibility that software alone may lack. Bennett notes robotics hasn’t had its “ChatGPT moment” yet and warns many robotics companies have raised large rounds without clear commercial promise, while A* focuses on seed-sized, task-specific bets.
- •Hardware/robotics expand AI beyond ‘brains in a jar’ into the physical world
- •Defensibility: sensors and real-world deployment are harder to replicate than software
- •Robotics hasn’t hit a breakout consumer/enterprise moment yet; commercialization still early
- •A* examples: vertical/scrappy robotics (Watney for data center cabling); edge AI/sensors
- 35:38 – 36:59
What’s next for A*: steady fund size, patient deployment, and founder proximity
They close with plans for continued focus rather than dramatic strategy changes: meet more founders, move quickly on seed leads, and remain patient on deployment. They also mention relocating offices and a founder-friendly culture of helping companies ‘move up a floor’ as they grow.
- •No major strategy shift: keep leading seed rounds and supporting founders over time
- •Modest fund growth: started at ~$300M; new fund not dramatically larger
- •Patience as a differentiator: not forcing capital deployment
- •Operational note: moving buildings; maintaining close proximity to founders
