The Twenty Minute VCSarah Guo: On Her New $101M Fund; How AI Impacts Inequality; AI Startups vs Incumbents | E1007
CHAPTERS
- 0:00 – 0:15
AI code generation’s dark side: malicious code at scale
A cold open sets the tone with a concrete AI risk: the same tools that make developers productive can also accelerate attackers. Sarah frames this as a near-term, practical security concern rather than a distant AGI scenario.
- •Code generation implies the ability to generate malicious code
- •Nation-states and hackers will adopt AI tooling quickly
- •Risk is about scale and accessibility, not novelty
- •Motivates a need for defenses and broader conversation
- 0:15 – 0:39
Reconnecting after six years and setting the stage for Conviction
Harry and Sarah reconnect, reflecting on their earlier conversation about conversational AI being “early.” They pivot to Sarah’s new firm, Conviction, and why the move matters now.
- •Context: last appearance was in 2017
- •Market timing lessons and how the landscape changed
- •Transition to discussing Conviction’s launch
- •Opens the question of leaving Greylock
- 0:39 – 1:42
Why Sarah left Greylock: early-stage focus and AI as a breaking change
Sarah explains that leaving Greylock wasn’t about dissatisfaction—it was about returning to the magic of zero-to-one and operating differently. The core catalyst: belief that AI represents a true platform shift requiring a new model of venture building.
- •Deep appreciation for Greylock’s people and platform
- •Desire to focus on early-stage (seed/Series A) work
- •Motivation to be an entrepreneur again and redesign the founder experience
- •AI is viewed as a “breaking change” driving the new firm
- 1:42 – 2:56
From GP to fund founder: existential dread, operations, and two customer sets
The conversation turns to what’s surprising about running a fund as a business. Sarah highlights operational realities—tools, hiring, strategy—and the dual responsibility to both entrepreneurs and LPs.
- •Running a fund is running a company (systems, payroll, office, recruiting)
- •More pressure and “existential dread” than being a partner at a large platform
- •Balancing service to founders with servicing LP relationships
- •Acknowledging the hidden work behind fund management
- 2:56 – 3:46
Conviction’s thesis: specializing in AI to win decades of value creation
Sarah lays out why Conviction is focused: she believes AI will create massive value and enable tiny teams to build billion-dollar companies. Specialization helps with selection, community access, and supporting founders with AI-specific needs.
- •AI as the biggest value creation opportunity of our lifetimes
- •Expectation of 10–20 person teams building billion-dollar outcomes
- •Specialist advantage: applied research, community building, matchmaking
- •AI-specific needs: model access, data, GPUs, design partners, safety/alignment
- 3:46 – 10:16
AI, abundance, and wealth inequality—plus the challenge of regulation
Harry raises concerns that AI could centralize wealth even as it boosts productivity. Sarah agrees inequality risks are real, but argues progress brings abundance—and society must address distribution through policy and democratic processes; regulation must catch up fast.
- •AI likely increases wealth inequality even while creating abundance
- •People adopt productivity tools quickly and don’t want them taken away
- •Distributional impacts require policy responses
- •Regulatory knowledge gaps exist; speed of change is the key difference
- •Builders and investors should partner with policymakers and educate
- 10:16 – 13:58
AI investing reality check: hype, pricing, and when big early rounds make sense
They unpack the “craziness” in AI funding: Sarah argues media amplifies a few extreme cases. She distinguishes between rare businesses that can deploy $100M well (often model training) versus the far larger set of capital-efficient applied AI companies.
- •AI hype is real but concentrated in a small number of headline deals
- •Character.AI cited as an example with exceptional engagement
- •Training frontier models is capital-intensive (teams + thousands of GPUs)
- •Fewer than ~10 cases justify training from scratch; most should apply/fine-tune
- •Startups should validate demand before scaling spend
- 13:58 – 17:18
Fund design choices: why a $100M fund, concentration, and avoiding “brand checks”
Harry challenges whether a smaller AI fund can compete in a world of huge rounds. Sarah explains why she intentionally stayed at $100M: constraints force focus, prevent structurally mismatched deals, and emphasize meaningful partnership over tiny “logo” allocations.
- •Sarah chose $100M intentionally despite ability to raise more
- •Constraints breed discipline for investors as well as founders
- •Conviction focuses on seed/Series A with more concentration than typical seed
- •Avoiding deals that can’t move the needle for fund outcomes
- •Long-term brand comes from returns, important companies, and founder reputation
- 17:18 – 23:00
Where AI is working now: legal AI, code generation, RPA rebirth, and multimodal creation
Sarah shares practical “low-hanging fruit” use cases where current error rates are tolerable and value is immediate. She highlights legal work (Harvey), the expanding frontier of code generation, automation, and multimodal creative tooling—plus what makes the right founding team.
- •Harvey as a prototypical ‘text-in/text-out’ legal application with clear ROI
- •Code generation is early; Copilot is just the starting point
- •RPA/workflow automation gets reinvented with LMs
- •Interest in LM tooling: retrieval, feedback, and infrastructure around models
- •Great teams often blend deep AI capability with strong domain/product insight
- 23:00 – 25:44
AI startups vs incumbents: speed, data moats, and who’s positioned best
They examine whether startups can beat incumbents when AI is moving fast. Sarah argues startups’ core edge remains speed; incumbents have advantages, but “data moats” are overstated—creativity and synthetic data can blunt that edge. Microsoft is called out as executing best via OpenAI; Apple/Amazon may need to invest more.
- •Startups’ main advantage is speed—more important in fast-shifting periods
- •‘Data moat’ narratives are often exaggerated; data is abundant/creatable
- •Outcome depends more on founders and execution than structural theory
- •Microsoft praised for strategic bets and market expansion attempts (e.g., search)
- •Amazon and Apple noted as needing more cutting-edge AI investment over time
- 25:44 – 30:32
Choosing what to build: idea generation, customer truth, and founder vs market nuance
Sarah argues great ideas rarely come from generic brainstorming; they come from high-resolution customer discovery or deep research questions tied to valuable markets. On founder vs market, she’s founder-first but insists founders must recognize market structure and have a credible plan to unlock distribution or shift constraints.
- •Generic ‘idea hunting’ yields generic companies; seek depth and uniqueness
- •High-resolution customer conversations reveal real priority and urgency
- •Example of research-driven opportunity: generating usable 3D models
- •Founders need uncomfortable exploration before committing to execution
- •Founder-first investing still respects market realities and distribution challenges
- 30:32 – 33:52
Learning from misses and rethinking defensibility at seed
Sarah shares a key regret: passing on iconic companies like Benchling and Rippling at pivotal moments—often because the market seemed unclear or founder risk felt scary. She then challenges the common seed-stage obsession with defensibility, arguing it doesn’t exist at the start; what matters is trajectory and the team’s ability to develop a defensibility thesis over time.
- •Anti-portfolio examples: Benchling and Rippling (missed due to perceived market/risk)
- •Greater willingness now to take founder-driven risk
- •Markets can change—and founders can reshape markets
- •Defensibility at seed is a flawed filter; there’s “no company yet”
- •Invest in trajectory, navigation ability, and evolving strategy
- 33:52 – 40:56
The future of venture: multi-stage scale vs boutiques, founder diligence, and the ‘generalist’ debate
They discuss why incentives and fund size shape behavior: large funds struggle to care deeply about small checks, while boutiques can be simpler and more aligned. Sarah urges founders to do investor references and understand internal dynamics, and she pushes back on absolute claims that generalist seed investing is dead—there are multiple ways to win.
- •Big funds face structural incentive challenges at small check sizes
- •Founders should choose partners based on who moves the needle in 18 months and long-term trust
- •Risks in large partnerships: groupthink, politics, seniority effects
- •Founders often underuse investor references as a diligence tool
- •Generalist vs specialist isn’t binary—different investor archetypes can succeed
- 40:56 – 54:29
Quickfire finish: macro outlook, AI services opportunity, funding models, AI abuse risks, LP culture, politics, and Conviction’s definition of success
A rapid set of topics closes the episode: Sarah expects continued tech ‘digestion’ after overcapitalization, and argues AI’s biggest market expansion may be in services, not just software. She also flags near-term security abuses (malicious code), explains Conviction’s limited reserves posture, critiques LP herd behavior, touches on 2024 politics, and defines success as best-in-class multiples, relevance, and being beloved by founders—while nudging AI toward productive, aligned outcomes.
- •Macro: tech faces a prolonged digestion period after ‘fat startup’ era
- •AI opportunity extends beyond software into services (legal as example)
- •Buy/short takes: pressure on undifferentiated subscale seed funds; skepticism on some growth approaches (e.g., Tiger)
- •Near-term AI abuse: malicious code generation and the need to invest in defenses
- •Reserves: early-stage focus means leaving later-round ownership on the table
- •LP world: clubby behavior and brand-following; desire for more independent thinking
- •Success for Conviction: top-tier multiples, relevance via important companies, small effective partnership, founder love, and aligned AI progress
