Dwarkesh PodcastBrett Harrison — FTX US former president speaks out
CHAPTERS
- 0:46 – 4:29
ETFs, passive investing, and the downside of index-driven correlations
Brett argues ETFs and passive investing are broadly beneficial for most investors because they lower costs and encourage diversification. The main systemic downside he flags is artificial correlation: stocks can move together more simply because they share index membership.
- •ETFs make diversified investing dramatically cheaper than in prior decades
- •Passive investing is generally appropriate for most individuals vs. stock picking
- •Index inclusion can create extra correlation unrelated to fundamentals (e.g., Tesla joining S&P 500)
- •Any price dislocations tend to be arbitraged away, but can persist temporarily due to liquidity/access frictions
- 4:29 – 6:07
Market-making strategies across Jane Street, Citadel, and the role of infrastructure
The conversation contrasts how major trading firms specialize in different niches, time horizons, and microstructure edges. Brett emphasizes that while trading strategies differ widely, the core engineering stack—connectivity, market data, simulation, and trader tooling—often looks similar.
- •Firms occupy distinct niches: ETFs, options, latency-arb, longer-horizon relative value
- •“HFT” is not a single strategy; horizons range from milliseconds to months
- •Shared needs: exchange connectivity, market data ingestion, simulation, visualization tools
- •Public perceptions often underestimate how different firms’ profit engines are
- 6:07 – 8:28
HFT engineering “hacks”: shaving latency and building 'good enough' models
Dwarkesh asks for finance equivalents of famous programming tricks; Brett gives examples from ultra-low-latency order entry. They discuss how HFT engineering resembles game graphics: optimize for speed and practical accuracy rather than theoretical perfection.
- •Example trick: pre-construct network messages and only fill the final bytes (e.g., price) at send time
- •Latency advantages come from both software and physical infrastructure (colo, microwave links)
- •Goal is “approximately right” trades fast, not perfect price models
- •Many micro-optimizations exist, but they’re rarely publicly visible
- 8:28 – 13:31
Is high-speed finance zero-sum? Price discovery, competition, and social value
Dwarkesh challenges whether nanosecond competition benefits society. Brett defends marginal competition as a path to tighter spreads and faster price discovery, while acknowledging it’s hard to define an 'optimal' speed limit for markets.
- •Competition at the margins tends to compress spreads and improve liquidity
- •Hard to set a principled cutoff: why would a millisecond be 'right' but a microsecond 'too fast'?
- •HFT spend is unlikely to be an outsized share of GDP; incentives imply value capture exists
- •Infrastructure built for trading (e.g., microwave networks) may have spillover benefits
- 13:31 – 18:39
Rare events, trader mental models, and how COVID changed trading operations
Brett describes how market-making trains probabilistic thinking about tail risks, but doesn’t necessarily translate into solutions for civilizational threats. He also recounts how top trading firms adapted operationally to COVID and remained profitable amid volatility.
- •Finance improves estimation of rare-event probabilities more than it improves societal problem-solving
- •COVID: financially flexible circles acted faster (e.g., leaving cities) once risk was salient
- •HFT/market-making firms proved resilient operationally despite remote-work/IP concerns
- •Volatility during COVID contributed to strong P&Ls for many firms
- 18:39 – 23:12
Interstellar markets, slow blockchains, and periodic batch auctions
They explore whether markets can function across huge latency distances, using DeFi as an analogy for slow settlement. Brett explains how AMMs and periodic auctions can reduce latency-based advantages, though real-world adoption has been mixed.
- •Interplanetary markets are technically possible but create severe adverse selection for distant traders
- •DeFi/AMMs show how slow systems can still support functioning markets via deterministic pricing rules
- •Periodic auctions can reduce latency advantages by prioritizing price over time within the batch
- •Some exchanges abandoned periodic auctions due to liquidity/price discovery complaints
- 23:12 – 32:07
Hiring as an inefficient market, plus Jane Street’s dev culture and language choices
The discussion pivots to hiring dynamics, emphasizing that labor markets are far less efficient than trading markets and can contain 'positive selection' opportunities. Brett also weighs OCaml vs. Rust, and explains why Jane Street’s internship pipeline is so lucrative and strategically valuable.
- •Hiring can reward exploring non-obvious talent pools; resumes from unknown schools may be positive signals
- •Strong brand/mission improves candidate selection by attracting top applicants
- •OCaml vs Rust: Rust could be preferable from scratch, but legacy OCaml infrastructure is massive
- •Internships: high pay reflects real work + conversion pipeline; training isn’t purely a cost
- 32:07 – 42:10
Quant culture vs Silicon Valley: pragmatism, creativity, and lifestyle
Brett contrasts quant finance’s pragmatic, profit-linked engineering culture with Silicon Valley’s emphasis on novelty and exploration. They also touch on the relatively inconspicuous consumption patterns of modern high-earning tech/finance workers and potential societal effects.
- •Quant firms optimize for direct profitability; SV culture often optimizes for novelty/creativity
- •Each culture could benefit from cross-pollination: creativity for quants, pragmatism for startups
- •Lifestyle tends closer to 'Walmart T-shirts' than Wolf of Wall Street in modern quant circles
- •Potential upside: younger wealth holders may seed startups, philanthropy, and dynamism earlier
- 42:10 – 47:08
How Brett met SBF: Jane Street, OCaml Bootcamp, and early impressions
Dwarkesh transitions to FTX, starting with Brett’s relationship with Sam at Jane Street. Brett describes Sam as smart and friendly, later increasingly contributing to systems design, and notably involved in early effective altruism/animal welfare circles.
- •Brett ran OCaml Bootcamp; Sam was an early student cohort
- •Sam was above average at Jane Street but not uniquely standout given the high bar
- •Their friendship strengthened via shared vegan/animal welfare interests
- •EA presence at Jane Street existed but wasn’t universal; Sam was known for high donation intent
- 47:08 – 51:19
Alameda’s early troubles and the road to FTX
Brett recounts intermittent contact with Sam after Jane Street, including a 2018 call about Alameda’s problems and possible pivots. In hindsight, he connects early infra issues, losses, and internal splits to the eventual evolution into FTX.
- •Sam told colleagues he’d join the Center for Effective Altruism, then started/expanded Alameda
- •Early Alameda issues: fragile Python scripting, operational mishaps, directional losses
- •Internal disagreement: rewrite vs incremental fixes; a faction split off
- •Brett first learned Sam became a billionaire through colleagues and media coverage
- 51:19 – 58:09
Joining FTX US: role ambiguity, goals for regulated products, and the Alameda 'wall' story
Brett explains why he left Citadel Securities for the opportunity to build FTX US, which was tiny compared to FTX.com. He also describes what he was told about Alameda: effectively independent, walled off, and treated like any other market maker—claims he later views with skepticism.
- •FTX.com was a top global exchange; FTX US was nascent with minimal volume
- •Ambitions: grow US spot, bring regulated derivatives onshore, potentially add stocks trading
- •FTX US and FTX.com were separate entities but shared/licensed core technology
- •Sam’s description of Alameda: self-running, walled off, no special privileges
- 58:09 – 1:09:16
Building FTX US and early accomplishments—then organizational trouble emerges
Brett highlights accomplishments: scaling a US team, regulated progress via LedgerX, and building a stocks platform largely himself. The chapter then turns to his core frustrations: inability to hire, centralized decision-making in the Bahamas, and an unsustainable codebase owned by a tiny dev clique.
- •US team grew to ~75–100; major work on compliance/legal/ops scaffolding
- •LedgerX/CFTC proposal: 24/7 margining, cross-collateralization, innovative clearing model
- •FTX US stocks platform: Brett claims he wrote most of the code
- •Major friction: Bahamas leadership blocked hiring and resisted a separate US dev org; 'call Gary/Nishad' culture
- 1:09:16 – 1:26:54
SBF’s leadership style: PR fixation, conflict avoidance, and internal frustration
Brett describes Sam’s leadership as largely absent day-to-day, with heavy emphasis on media, branding, and high-profile deals. He explains how SBF’s accessibility created a 'flywheel' of reputation laundering—investors, celebrities, and media reinforcing one another—despite internal dysfunction.
- •SBF prioritized PR/visibility over internal management and timely decisions
- •Employees felt directionless; some simultaneously hero-worshipped Sam
- •Reputation flywheel: celebrity endorsements + investor praise + media access created inflated credibility
- •Brett’s direct criticism in the Bahamas triggered an unusually angry reaction from Sam
- 1:26:54 – 1:36:23
Alameda proximity, red flags in hindsight, and the late bonuses as a liquidity signal
They discuss the perceived separation between Alameda and FTX, including physical office separation and the normality of 'Chinese walls' in finance. Brett’s most concrete hindsight red flag is delayed mid-2022 bonuses, which he now interprets as possible liquidity stress; otherwise, many issues looked like 'bad management' rather than fraud at the time.
- •Brett did not see evidence of special 'hard-coded' Alameda privileges in the codebase
- •Physical separation between Alameda and exchange offices seemed standard, like different floors in banks
- •Key hindsight red flag: unusually late mid-2022 bonuses amid broader market stress
- •Other red flags: concentrated code ownership, blocked hiring, lack of US governance separation
- 1:36:23 – 1:45:44
The resignation letter, internal threats, and SBF’s attempt to control the narrative
Brett details the ultimatum letter: weekly communication, US leadership autonomy, and expanding/redistributing the tech org away from a two-person bottleneck. He describes SBF’s rejection, subsequent intimidation (including demands for an apology letter), and later efforts to portray Brett as fired and unwilling to move to Miami after office closures.
- •Letter demands: regular 1:1 communication, separate US C-level leadership, grow/distribute engineering authority
- •SBF dismissed calls as inefficient, defended the dev setup, resisted delegating titles/authority
- •A deputy threatened reputational retaliation and bonus withholding; demanded an edited apology letter
- •After Brett left, SBF announced office closures and told outsiders Brett was partly fired/constructively fired
- 1:45:44 – 2:04:31
FTX collapse from the outside: disbelief, fundraising freeze, and lessons on oversight
Brett recounts learning about the crisis via Twitter and the Binance acquisition tweet, describing shock because he still held funds and equity. He then reflects on why it was hard for outsiders and employees to detect fraud given falsified records, and argues for stronger governance (e.g., independent boards) plus enforcement rather than pure 'never trust anyone' paranoia.
- •First inkling: social-media panic about FTT/Alameda; Binance acquisition tweet triggered disbelief
- •Architect fundraising paused as investors triaged exposure and broader contagion
- •Brett says he lacked access to FTX.com financials; even audited info was allegedly manipulated
- •Proposed mitigations: mandated independent boards/governance and regulatory oversight; enforcement is essential
- 2:04:31 – 2:15:45
Regulators, the LedgerX proposal, and why crypto needs rules (not just arbitrage)
They discuss regulators’ pre-collapse view of crypto as viable but needing an appropriate regulatory envelope. Brett defends FTX’s CFTC proposal as a meaningful modernization attempt (24/7 markets) while acknowledging concerns about liquidation cascades and role-consolidation; he rejects the idea that crypto’s 'point' is regulatory arbitrage.
- •Regulators viewed FTX as unusually collaborative compared to offshore/avoidant players
- •LedgerX model combined DTC clearing, cross-collateralization, and auto-liquidation—novel in its combination
- •Concerns: systemic risk from liquidation cascades and consolidating roles normally split across entities
- •Crypto regulation is inevitable; working with regulators helps create durable institutional adoption
- 2:15:45 – 2:30:08
Architect.xyz: unified crypto trading infrastructure and where crypto markets may go next
Brett introduces Architect as a connectivity and tooling layer across centralized exchanges, DeFi protocols, and custody options, reducing duplicated engineering effort for traders and institutions. He explains why exchanges won’t build cross-venue tools themselves, how FTX taught him API design and trust minimization, and why he’s building for multiple possible futures (CeFi, DeFi, or TradFi incumbents entering).
- •Architect aims to unify market data, execution, custody, and DeFi access via deployable commodity software
- •Crypto’s lower infrastructure barriers enable sophisticated individuals as well as institutions to trade
- •Exchanges’ incentives discourage building tools that make it easy to route flow to competitors
- •Post-FTX design focus: don’t require users to hand over private credentials; wrap exchange idiosyncrasies with a clean API
- 2:30:08 – 2:37:37
Institutional interest, use cases of crypto, and on-chain settlement for real-world assets
Brett argues institutional interest remains strong despite lower volumes post-FTX, with many banks still pursuing blockchain initiatives. They close on the social value of efficient crypto markets (as with commodities), the possibility of tokenized securities trading on-chain, and the motivation for blockchain-based settlement as an upgrade over error-prone legacy systems.
- •Institutions still trade and invest; many are waiting for clearer US regulation and better tooling
- •Value case depends on token utility: store of value, gas for decentralized computation, or security-like funding
- •On-chain settlement could reduce T+2 delays and operational errors common in equities/fixed income
- •Crypto markets may be less efficient than equities, leaving room for informed traders beyond incumbents
