A Conversation with Robinhood co-founder and CEO Vlad Tenev

KEYPOINTS

• 01

  The GameStop restriction was a one-day risk decision that a viral good-vs-evil story turned into 'collusion.'

  Vlad Tenev says an automated overnight collateral file arrived with huge, shifting numbers in a situation with no precedent. Robinhood put GameStop and other names on 'position closing only' for about a day. In normal times it would have been an innocuous risk call, but the narrative of retail investors taking down hedge funds — and a company literally named Robinhood — made the false story irresistible.

• 02

  A juicy falsehood is more powerful than a boring truth, and facts rarely dislodge a story once it has mind share.

  Robinhood had no meaningful business with hedge funds, yet the collusion story stuck. Another rumour — that the White House called Sequoia to pressure Robinhood — had thousands of Reddit reposts and Sequoia had to refute it. Shane adds that people can hold such stories for 20 years.

• 03

  2022 was harder than GameStop because every tailwind reversed slowly at once.

  Stimulus ended, inflation rose and rates jumped to the highest in over 30 years. With roughly 7% expected stock returns versus about 5% in cash, first-time investors shifted to cash. Robinhood fell from a ~$32B IPO valuation in 2021 to trading around $6, down 80%+, called a 'broken IPO,' with advice to go private.

• 04

  Refusing to be an ostrich: build for the environment you're actually in.

  While some mortgage firms waited for rates to fall, Robinhood asked what would help customers thrive with high rates. It revived Gold around top yield on uninvested cash, then launched Robinhood Retirement (1.5M+ accounts, 3% match for Gold members). The company now has 11 business lines each generating over $100M a year.

• 05

  COVID made Robinhood 'bulk up' with fat, so 2022 became a forced lean-out.

  It went from ~700 people and ~$200M revenue at end-2019 to thousands of people and close to $1B a year later — growth built on inputs that weren't sustainable ('you only have one COVID'). Remote hires had never met. In hindsight he'd have enshrined values earlier.

• 06

  Practice reversing decisions on small things so big reversals feel less scary.

  Undoing a decision means admitting you were wrong, so leaders often half-undo it and get no result. Tenev killed COVID-era 'wellness days', when whole teams took the same day off, pointing staff to generous PTO. People feared revolt; there was one day of complaints and then silence.

• 07

  Pay for impact, not org size, so you don't reward empire building.

  Values are high performance, safety always, lean and disciplined, and 'one Robinhood.' The job is pitched as doing in one year what another company expects in ten. If a hire clearly isn't working, even at three weeks, they part quickly; six months is too long once you know, though slow starters with rare strengths can get time.

• 08

  Hire early-career people so the company doesn't age out of its customers.

  Robinhood recruited heavily near Stanford. Tenev argues brokerages become generational — Schwab for boomers, E*Trade for Gen X — as staff age. Interns ship production work; he notes he himself 'started as an intern' straight out of school.

• 09

  AI adoption only works if you track the right metrics in the highest-leverage teams.

  Robinhood focused first on customer service (AI deflection rate) and engineering (share of commits from AI, checked against total commits per engineer). Teams now budget compute alongside headcount. He sees creative and marketing — with tools like ElevenLabs, Midjourney, Runway — as the next frontier, with 100–1,000× more personalised output.

• 10

  Math-trained AI can be verified, which lets it generate its own training data.

  Tenev chairs Harmonic, whose agent-network model Aristotle got gold-medal performance at the International Mathematical Olympiad (5 of 6 problems) and helped solve an open Erdős problem. Formalising proofs as code allows machine checking, so proven lemmas feed back into training — closer to chess AI than a chat model. He expects specialised models, since domain users generate domain data.