The Twenty Minute VCInside Sequoia's Investment Committee | How the SpaceX & Citadel Deals Went Down | Julien Bek
CHAPTERS
- 0:00 – 4:37
Julien’s values and family responsibility (and why it shapes how he works)
Julien shares the personal story of caring for his father after a neurological condition emerged in Julien’s early 20s. Harry and Julien discuss values, duty, and how doing things for others can be a powerful motivator—setting the emotional foundation for the rest of the conversation about work and investing.
- •Julien’s role as sole caregiver (no siblings) and how that formed his mindset
- •Family background and early-life influences that shaped his values
- •Harry’s pushback on “do it for you” advice; motivation through responsibility
- •Values as a source of clarity in hard decisions
- 4:37 – 7:02
What people misunderstand about Sequoia: “hunters,” not inbound deal tourists
Julien explains that the common perception—Sequoia just waits for elite founders to call—is wrong. He describes Sequoia’s culture as a sports team where everyone is expected to perform, compete, and win together while behaving exceptionally as individuals.
- •Sequoia culture: competitive “sports team” with constant performance expectations
- •Small early-stage team dynamics (like a football team) and shared accountability
- •Winning as a team while maintaining high individual standards
- •Misconception that brand alone delivers the best deals inbound
- 7:02 – 8:28
How Sequoia won Citadel Securities: relationships measured in decades
Using the Citadel Securities investment, Julien illustrates Sequoia’s long-game approach. The deal came from Constantine Bühler building a relationship with Ken Griffin since he was a student and persistently asking over many years.
- •Citadel’s uniqueness: historically no outside capital
- •Constantine’s relationship-building over years as the decisive edge
- •Persistence as strategy: repeatedly asking until the timing is right
- •Sequoia’s “hunt” is often relationship-driven, not purely process-driven
- 8:28 – 11:05
Revisiting priors and paying up later: the logic behind the $2.5B Anthropic bet
Harry and Julien dig into Sequoia’s willingness to update its thinking and invest later at a much higher price when the world changes. Julien frames this as revisiting priors—especially important in AI, where outcomes follow exponential curves humans systematically underestimate.
- •“Revisiting priors” as a core investing discipline
- •Humans are bad at exponentials; AI forces uncomfortable re-calibration
- •Ego management: investing later at a higher price after earlier skepticism
- •Power-law dynamics and the new scale of venture outcomes
- 11:05 – 12:11
Neo Labs skepticism: why “the next lab” can look like Quora vs Facebook
Julien offers a contrarian take on Neo Labs-style companies: many resemble strong products that still lose to a platform shift. He argues that backing a truly novel lab today requires an ‘N-of-1’ founder pursuing a fundamentally different architecture—not “same thing, better.”
- •Neo Labs compared to Quora/StumbleUpon in the shadow of a Facebook-scale shift
- •Requirement for novelty: different architecture vs incremental improvement
- •Backing ‘N-of-1’ founders as the viable strategy in crowded lab landscapes
- •Example: Ineffable and the bet on unique research direction
- 12:11 – 13:35
Is Series A the hardest today? Sector-dependent goalposts (bits vs atoms)
Julien challenges the blanket claim that Series A is universally hardest. In hardware/physical AI, validation takes longer and requires different milestone thinking (prototype progress vs revenue), likely forcing more collaboration among investors to fund longer development cycles.
- •“Hardest stage” depends on sector, especially hardware vs software
- •New milestones: prototype/validation rather than revenue ramps
- •Physical AI is capital- and time-intensive (“moving atoms, not bits”)
- •Prediction: more investor collaboration to finance longer arcs
- 13:35 – 27:06
Why Julien won’t compromise on ownership: time is the real constraint
Julien argues he stays ownership-centric not because outcomes are smaller, but because partner time and depth of engagement are scarce. His model is to partner intensely with a few founders per year, acting like a co-founder in the passenger seat, which only works with meaningful ownership and focus.
- •Career constraint: ~20 deep board-level partnerships vs hundreds of tiny stakes
- •High-touch operating support (customers, hires, GTM) requires concentration
- •Ownership as a reflection of time commitment and accountability
- •Contrasts with high-volume angel-style investing models
- 27:06 – 34:24
Inside Sequoia’s IC: async memos, founder pitches, voting—and “front stabbing”
Julien explains how Sequoia’s investment committee works, including experiments with asynchronous feedback to combine ‘slow thinking’ with traditional IC debate. Founders still pitch the full group; partners vote (pre- and post-discussion), and the sponsor can still proceed—at reputational risk if wrong.
- •Monday IC tradition, now evolving with asynchronous written input
- •Founders pitch the full early-stage partnership (small group, high intensity)
- •Voting as signal: pre-discussion pulse + post-discussion decision vote
- •Sponsor autonomy exists, but conviction must be earned and defensible
- •Culture of directness (“front stabbing”) and heated but functional debate
- 34:24 – 39:42
Founder reading as a superpower: vulnerability, ‘ask why,’ and spotting fraud
Julien shares his approach to reading founders quickly: lead with vulnerability to elicit authenticity, then probe with repeated ‘why’ questions to test coherence. He recounts uncovering a fraudulent founder by noticing inconsistencies, body language, and evasive behavior, and then warning other investors.
- •Vulnerability as a tool to unlock founder openness and real signal
- •30-minute mandate: identify the founder’s ‘spike’ and potential exceptionalism
- •“Ask why five times” to reveal truth, logic, and authenticity
- •Fraud case: suspicious story, escalating nervousness, canceled meeting, investor alerts
- •Omission errors (missing great companies) are more costly than commission errors
- 39:42 – 40:36
Arrogance, spikes, and the Don Valentine 2x2: liking founders vs making money
Julien argues arrogance is not inherently a red flag if it’s the cost of a genuine spike. Referencing Don Valentine’s 2x2 (founders you like vs founders who make you money), he emphasizes the job is to find where returns come from, not to optimize for personal preference.
- •Don Valentine 2x2: ‘likable’ is different from ‘profitable’
- •Arrogance can be functional when tied to real exceptional ability
- •Key diagnostic: is arrogance masking weakness or supporting a spike?
- •Investor task: understand what traits predict outcomes, not comfort
- 40:36 – 46:12
Founder misreads and cultural calibration: Lovable, country context, and NPS bias
Julien admits he misread Anton Osaka (Lovable) because he came unprepared and didn’t ask intentional questions. The discussion expands into how founder style and customer feedback vary by country—e.g., German customers giving ‘7’ can be a strong signal—requiring cultural calibration in diligence.
- •Lovable miss: lack of meeting plan and insufficient probing led to underestimation
- •Cultural communication styles (e.g., ‘thoughtful Swede’ vs ‘aggressive Swede’)
- •Reference calibration: Germans/French often underrate; Americans may overrate
- •Importance of intentional questioning and structured founder evaluation
- 46:12 – 54:00
Mentors at Sequoia: Doug, Pat, Alfred, Shaun—reference frameworks and judgment
Julien shares one key lesson from each leader: Doug’s ‘best and worst reference’ interview question; Pat’s vector model (direction × magnitude); Alfred’s warning not to confuse outlier operators with outlier founders; and Shaun’s expanded model of human capability (IQ/EQ plus judgment and political coefficient).
- •Doug: ask ‘best reference’ then ‘worst reference’ to reveal self-awareness and texture
- •Shaun: ELO theory—outliers best identify other outliers; references need exceptional calibrators
- •Pat: vectors = direction (motivation) × magnitude (ambition/endurance)
- •Alfred: gold-plated CVs can signal operator excellence, not founder outlier status
- •Shaun: judgment > IQ; political coefficient (PQ) > EQ in complex systems
- 54:00 – 1:01:46
Agents as the new customer: ‘bits-perfect’ conversion, bias, and the AEO economy
Julien presents a thesis that agent traffic will rival and then dwarf human traffic, requiring businesses to optimize for agent decision-making rather than human UI. He argues the ‘UI goes to zero’ view is too simplistic because agents have biases shaped by training and post-training, creating a new economy with categories like AEO (answer engine optimization).
- •Agent traffic approaching parity; potential 1000× agent-to-human in future
- •Shift from pixel-perfect UI to ‘bits-perfect’ platforms that convert agents
- •Agents have biases; brand/distribution still matters through model behavior
- •AEO as an early category; example: Profound for chat-interface visibility
- •Parallel agent economy creates new markets and new measurement/optimization tools
- 1:01:46 – 1:06:16
Apps vs infrastructure, switching costs, and margins in an AI transition phase
Harry and Julien debate durability: Harry prefers infrastructure certainty, while Julien argues Sequoia can hold opposing truths and invest in both layers. Julien explains why switching costs and trust (data gravity, enterprise controls) preserve margins in many contexts, and why frontier-model dependence will fade for many human-facing apps as open models become “good enough.”
- •Investing across stack: infrastructure winners now; application stickiness compounds later
- •Switching costs remain real in enterprise contexts despite agent agility
- •Margins: temporary pressure while operating at the frontier; optimization later with open weights
- •Machine-to-machine workflows may sustain frontier demand longer than human-facing apps
- •Example signals: Rillet’s expansion beyond tech (“Project Iowa”) into the real economy
- 1:06:16 – 1:16:55
Outcome-selling ‘software that masquerades as services’—and why Julien won’t back bolt-on AI
Julien clarifies his viral ‘services are the next trillion-dollar economy’ thesis: the trillion-dollar winner will sell outcomes (autopilot) while retaining software-like margins as AI replaces most labor over time. He cites customer support as already outcome-priced (e.g., Sierra), but argues against traditional services firms “bolting on” AI because they struggle to attract frontier talent and to harness the right data.
- •Copilot → autopilot shift enables selling outcomes instead of tools
- •Economics: capturing the $6 services spend vs competing for the $1 software spend
- •Customer support as an early autopilot category; outcome pricing per ticket resolution
- •Human judgment remains, but the mix shifts from lots of humans to lots of AI
- •Skepticism of ‘services first, software later’ for venture-scale outcomes: talent and data constraints
- 1:16:55 – 1:28:27
Quick-fire and closing: category bets, intensity, Trade Republic miss, and Revolut’s lesson
In quick-fire, Julien calls legal overfunded, BCIs underfunded, Cursor a standout agent company, and intensity the non-negotiable founder trait. He shares a personal investing story: missing Revolut professionally early in his career, then investing personally with his mother—who ultimately retired after a massive multiple—before closing with optimism about AI’s impact on life sciences.
- •Overfunded: legal (too many me-tos); Underfunded: brain-computer interfaces
- •Best agent company outside Sequoia: Cursor; best angel: Gloria (Puzzle)
- •Haunting miss: misunderstanding Trade Republic vs Revolut competitive dynamics
- •Founder trait he won’t compromise on: intensity
- •Revolut story: conviction, persistence, personal capital constraints, and long-term holding
- •Biggest excitement: AI enabling breakthroughs in biology and disease cures
