CharlieKey takeaways from your favorite podcasts
← 22 October 2026Masters of Scale

OpenAI CFO Sarah Friar & Reid Hoffman on where AI goes from here | Masters of Scale Summit 2026

Sarah Friar argues AI is moving from answering to doing, and that OpenAI must earn trust by pacing the frontier, showing real outcomes in health and security, and asking communities what they need.

Watch on YouTube ↗

Charlie's summary

  1. AI is moving from asking to doing: everyone can now have a virtual coworker, and 90% of users get it free

    Sarah describes a shift from the ChatGPT moment of asking to the agent moment of doing. 1.2 billion people use ChatGPT weekly, “a fifth of the world's population.” She says 90% of them get AI free because ads (100,000 advertisers, 89 countries) fund a durable business model. The mission is “AGI for the benefit of humanity, not AGI for the benefit of humanity who can pay for it.”

    ▶ 1:03
  2. Defenders need frontier tools now: AI cyber capability is rising and attacks are getting physical

    Sarah says OpenAI used an N minus one model to find vulnerabilities in PHP, which underpins about 70% of websites, and worked in secret to get them fixed before attackers found them. She cites hackers in France raising freezer temperatures at a supermarket chain to spoil goods. Her view is that bad actors already have powerful Chinese models, so defenders must get frontier tools and patch quickly.

    ▶ 3:38
  3. Pacing the frontier: OpenAI holds back its most powerful model until safety work catches up

    Sarah says “pacing the frontier” means giving researchers and the world time to digest the implications. OpenAI is holding back its strongest model, which solved the Navier-Stokes problem, until safety and alignment are ready. She admits this isn't great for business short term but calls it mission-aligned and good long term. She links theory to practice: fluid dynamics gains could mean more efficient heart valves and data center cooling, and so less water use.

    ▶ 5:10
  4. Trust is earned by filling the bucket, and healthcare outcomes are where OpenAI starts

    Sarah says trust “doesn't just fall down on you like mana from heaven”; it is grindy, like filling a bucket. About a third of weekly users ask a health question, and in health deserts like West Texas it can reach 80-90%. At Boston Children's, models have resolved 40 of 300-plus unsolved diseases. An HCA proof of concept cut patient hospital time by 25%. Showing outcomes beats saying trust us.

    ▶ 8:46

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.