Dwarkesh PodcastAgustin Lebron - Trading, Crypto, and Adverse Selection
CHAPTERS
- 0:00 – 3:29
Why most people shouldn’t trade + Agustin’s path from engineering/poker to Jane Street
Dwarkesh opens with a “don’t trade” reading of Agustin’s book, and Agustin largely agrees—trading is harder than people think and often not the best way for a smart person to make money. Agustin then shares his background: engineering, online poker, joining Jane Street in 2008, then consulting and starting a crypto company.
- •Trading is a brutally competitive job; most people lack true edge
- •If you can succeed in trading, you could likely find easier ways to succeed elsewhere
- •Agustin’s career arc: chip design → poker → quant trading (Jane Street) → consulting → crypto startup
- •Early framing of the conversation around incentives and “edge”
- 3:29 – 7:38
Adverse selection beyond markets: hiring, motivation, and “why are you applying?”
The discussion uses hiring as an analogy for trading adverse selection: employers face selection problems both in who applies and who accepts offers. Agustin explains why certain stated motivations (e.g., “I’ve always wanted to be a trader”) can be a red flag, and what healthier motivations look like.
- •Hiring markets are inherently adverse-selection-heavy for employers
- •The applicant pool is selected; the best employees are often retained by current employers
- •Employees choose among offers, creating additional selection pressure
- •Pure money-maximization can correlate with poor traits; curiosity and love of the game matter
- 7:38 – 8:51
How top trading firms evaluate potential (and why retail trading experience can hurt)
Dwarkesh asks how firms identify people who will thrive in trading if prior domain experience isn’t valued. Agustin describes looking for evidence of competitive obsession and intrinsic drive—like mastery of obscure games—rather than surface-level trading credentials.
- •The “third-best in a weird chess variant” as a signal of grind + intrinsic motivation
- •Competitive enjoyment and willingness to guess/iterate are key traits
- •Retail trading can create misconceptions that must be unlearned
- •Blank-slate, high-aptitude hires can be easier to train than “experienced” but miscalibrated ones
- 8:51 – 11:22
Why trading domain expertise is overrated + what 6–18 months of trader training looks like
Agustin argues that “domain expertise” is often misunderstood: retail trading is a different domain from professional market making. He explains what firms actually do during the long ramp-up period—structured learning, feedback loops, and repeated decision practice.
- •Retail trading ≠ professional trading; the mental model mismatch is costly
- •Many personal-account traders are likely negative-edge; not realizing that is itself a bad sign
- •Training is apprenticeship-like: continuous probing of reasoning and decisions
- •Bootcamps and structured curricula accelerate learning compared to pure osmosis
- 11:22 – 14:53
Talent arbitrage in tech: screen for ability, not legible skills
Agustin generalizes trading-style hiring to tech: companies overweight easy-to-measure skills and underweight potential. He outlines a vision for global mass screening and bootcamps to unlock under-tapped talent, especially in developing countries.
- •Common tech hiring failure: selecting for legible skills rather than ability/potential
- •Bootcamps can convert raw aptitude into job-ready capability
- •Big opportunity in identifying top 0.1% talent globally and bridging them to Western jobs
- •Selection mechanisms could look like “games” that reward persistence and learning
- 14:53 – 19:26
Being on the wrong side of adverse selection: signaling, interviews, and why brainteasers exist
Dwarkesh asks how individuals can overcome adverse selection when they’re the “good risk” being pooled with lemons (insurance, hiring). Agustin emphasizes evaluating interviewers and role fit, then explains why brainteasers persist: they often serve as IQ proxies under legal constraints, though they’re frequently misused.
- •Use mechanisms that credibly signal quality (e.g., telematics for safe driving)
- •Job seekers should evaluate interviewers to infer real job demands and culture
- •Status heuristics can cause people to adverse-select themselves into poor fits
- •Brainteasers often approximate general intelligence tests; quality depends on interviewer skill
- 19:26 – 21:17
Startup hiring reality: the ‘A-player’ delusion and barbell team construction
Agustin critiques the common startup desire to hire only elite candidates despite lacking the brand/pay to close them. He proposes a pragmatic barbell strategy: pay for a few truly top people and build the rest of the org with more junior/lower-cost hires placed in well-scoped roles.
- •Many startups build hyper-selective processes but can’t land the candidates they target
- •Resulting failure mode: hiring mediocre people who merely present well
- •Better approach: a few 90th-percentile anchors + many scoped, cost-effective contributors
- •Clarity about what roles truly require top talent improves organizational ROI
- 21:17 – 25:07
Invest in training (and don’t get fooled by ‘poached’ talent pools)
The conversation turns to concrete hiring arbitrage: hire more junior and invest in training rather than overpaying for the 2–3 year experience “kink.” Agustin gives an example of firms poaching from Intuit while missing the fact that the best employees are retained—classic adverse selection.
- •Best hiring arbitrage often comes from going more junior
- •The 2–3 year salary jump reflects market distortion and underinvestment in training
- •Training lets you shape employees to your needs and increases retention
- •Poaching pipelines can be adverse-selected: you see who is available, not who is best
- 25:07 – 28:47
Is finance too big? Regulation, deadweight loss, and what’s worth banning
Dwarkesh asks whether finance’s ~9% share of GDP is socially excessive. Agustin is ambivalent: much of the work is zero-sum, but markets haven’t found a clearly better mechanism; regulation is also complex with trade-offs, and some products may be actively harmful for retail.
- •Finance feels zero-sum from the inside, but outside-view evidence is mixed
- •Familiarity can create legitimacy bias for insiders
- •Possible policy lever: restrict retail access to leveraged ETFs/volatility products
- •Regulatory complexity (e.g., capital rules) can be a deadweight loss but simplification risks stability
- 28:47 – 31:31
Future of trading firms: service value, volatility, consolidation, and economies of scale
Agustin argues market makers persist because they provide a real service, especially in volatile times when others won’t take risk. He predicts continued consolidation due to economies of scale, while new frontiers (like crypto) periodically reset the landscape and then consolidate again.
- •Market makers earn by taking risks others avoid; volatility creates outsized opportunities
- •Even efficient markets still need liquidity and risk warehousing
- •Economies of scale push toward consolidation over time
- •New markets (crypto) start as ‘Wild West’ and then consolidate over a decade
- 31:31 – 35:46
AI in trading: what automation helps, and why ‘just throw ML at prices’ fails
They discuss how technology already expanded traders’ reach (10 stocks to 100+) and will continue to automate model testing and workflows. Agustin is skeptical that scaling LLMs alone yields robust trading intelligence; key issues include sample inefficiency and lack of true semantic understanding, plus the reflexive/agentic nature of markets.
- •Tech increases efficiency in human time; traders can manage more markets than before
- •LLM scaling may not produce the kind of grounded understanding trading demands
- •Markets are reflexive: profitable signals attract competitors and decay
- •Backtests can mislead due to execution/adverse selection and strategic response by others
- 35:46 – 39:23
Rationality/EA and Jane Street: shared norms and the culture fit
Dwarkesh notes the overlap between Jane Street and rationality/EA communities. Agustin attributes it to shared emphasis on not fooling yourself, quantitative thinking, and a collegial culture that suits people who enjoy collaborative truth-seeking.
- •Shared ethos: ‘shut up and multiply’ / disciplined belief-updating
- •Trading requires modeling other agents and incentives, not just static systems
- •Jane Street’s collegiality amplifies the appeal for rationalist/EA-aligned people
- •Cross-pollination between communities and industry practice goes both ways
- 39:23 – 48:31
Software, technical debt as finance, and engineering choices in high-stakes systems
Dwarkesh asks about technical debt through a financial lens and how finance firms make engineering trade-offs. Agustin discusses why startups rationally take ‘non-recourse’ tech debt, how big companies become debt-servicing machines, and why Jane Street’s OCaml/tooling choices emphasize correctness and safety—while reminding that software is largely sociology and complexity management.
- •Startups should take technical debt strategically (MVPs), expecting rewrites if successful
- •Large firms accumulate ‘software archaeology’ and spend heavily maintaining old stacks
- •Strong static typing (OCaml) helps make impossible states unrepresentable
- •Engineering in trading must avoid catastrophic ‘hot loop’ failures; culture/process matter most
- 48:31 – 57:03
Crypto market structure: why AMMs can lose, what success looks like, and real-world use cases
Agustin explains why trading against Jane Street still happens: counterparties pay for convenience and execution, not because they expect to outsmart market makers. He then critiques CFMM/AMM liquidity provision as structurally adverse-selected, contrasts it with equity risk premia, and lays out a ‘success case’ for crypto as integration into financial plumbing with clear wins like remittances and credentialing.
- •Jane Street profits from providing liquidity/services to end-users, not other market makers
- •AMM LPs face adverse selection; fixed fees may not compensate under changing conditions
- •Passive equity investing differs from passive LP’ing because equities have productive cash flows
- •Crypto’s success = hybrid integration with TradFi; wins include remittances, FX friction reduction, portable credentials (NFT-like)
- 57:03 – 1:04:40
Book recommendations + career advice: ‘life is long’ and sequential deep expertise
Agustin recommends three books spanning software, risk, and human nature. He closes with advice for young people: optimize for long horizons, build deep expertise in phases, and recognize different discovery styles (evolutionary vs revolutionary) that may suit different personalities.
- •Recommended books: *A Deepness in the Sky*, *Red-Blooded Risk* (or *Poker Face of Wall Street*), *Kolyma Stories*
- •Core advice: treat life as long; you have time for multiple skill arcs
- •“Sequential excellence”: go deep for years, then pivot and reuse perspective
- •Different creators: boundary-pushers vs field-inventors; self-awareness matters
