OpenAI’s Head of ChatGPT: We’re entering a new era of AI (again) | Tibo Sottiaux
Thibaut Sottiaux, Head of ChatGPT at OpenAI, discusses the launch of Dots (AI agents), the future of work with AI, and how builders should prepare for a world where most internet actions are performed by agents. He emphasizes the importance of safety, community-driven development, and the evolution from configuration-heavy systems to naturally learning AI assistants.
Summary
In this interview, Thibaut Sottiaux discusses OpenAI's latest product launches, particularly Dots—AI agents designed to work 24/7, learn user preferences, and operate autonomously across multiple devices. He explains that Dots represent a shift away from manual loop configuration toward systems that learn naturally from feedback and user goals. Sottiaux describes his own workflow of expanding and contracting teams of agents based on model capabilities, noting that larger model breakthroughs allow consolidation of agent teams.
On the broader future of AI, Sottiaux emphasizes three unprice-in factors: most internet actions will be taken by agents, models will become exponentially cheaper and faster, and modalities will integrate seamlessly. He argues that builders are not yet thinking ambitiously enough about this 10x improvement trajectory over the next year.
Regarding safety and alignment, Sottiaux stresses OpenAI's investment in monitoring, guardrails, and security systems alongside primary model development. He notes they deliberately did not release a more capable model (6.1 Astra) and are comfortable with the current Astra release level. He frames this as both a responsibility and an aligned incentive—OpenAI cannot afford to fail with 1.2 billion users.
On engineering and hiring, Sottiaux highlights that fast typing is becoming obsolete while taste, understanding users, and founder mentality are increasingly valuable. He mentions OpenAI has over 120 ex-founders on staff. He also discusses the ecosystem strategy, including revenue sharing with popular plugins and partnerships with companies like Notion.
Sottiaux addresses the human element, acknowledging concerns about loneliness and context-switching with agents. He emphasizes the ideal future involves AI integrated into physical spaces as conversational partners rather than screen-based interactions. He expresses his own excitement about rediscovering the engineering flow state through faster models and voice interaction.
Finally, he reflects on changes to his own beliefs: reaching current model capabilities happened faster than expected, the importance of voice interaction was underestimated, and younger workforce entrants have an advantage in embracing AI-first approaches.
About this episode
<p><strong>Tibo Sottiaux</strong> leads ChatGPT and Codex at OpenAI. Under his stewardship, OpenAI has shipped some of its most consequential consumer launches: Codex, ChatGPT Work, and most recently, the new Dots personal agent platform. He joined me at DevDay, hours after his team launched more than 20 new products, to talk about where things are heading.</p><p></p><p><strong>We discuss:</strong></p><p>1. Dots, OpenAI’s new personal AI assistant</p><p>2. Why most actions on the internet will soon be taken by agents</p><p>3. Why loops, graphs, and fine-tuning agent workflows are a passing phase</p><p>4. Which skills are trending up and down in the AI era</p><p>5. How Tibo’s Dot warned him about a production outage five minutes before the launch</p><p>6. OpenAI’s approach to AI safety</p><p>—</p><p><strong>Brought to you by:</strong></p><p><a href="https://workos.com/lenny" target="_blank"><strong>WorkOS</strong></a>—Make your app enterprise-ready, with SSO, SCIM, RBAC, and more</p><p><a href="https://getdx.com/?utm_source=lenny&utm_medium=sponsorship&utm_campaign=q42026" target="_blank"><strong>DX</strong></a>—Engineering intelligence for the AI era</p><p>—</p><p><strong>Episode transcript:</strong> <a href="https://www.lennysnewsletter.com/p/openais-head-of-chatgpt-were-entering" target="_blank">https://www.lennysnewsletter.com/p/openais-head-of-chatgpt-were-entering</a></p><p>—</p><p><strong>Archive of all Lenny's Podcast transcripts: </strong><a href="https://www.dropbox.com/scl/fo/yxi4s2w998p1gvtpu4193/AMdNPR8AOw0lMklwtnC0TrQ?rlkey=j06x0nipoti519e0xgm23zsn9&st=ahz0fj11&dl=0" target="_blank">https://www.dropbox.com/scl/fo/yxi4s2w998p1gvtpu4193/AMdNPR8AOw0lMklwtnC0TrQ?rlkey=j06x0nipoti519e0xgm23zsn9&st=ahz0fj11&dl=0</a></p><p>—</p><p><strong>Where to find Tibo Sottiaux:</strong></p><p>• X: <a href="https://x.com/thsottiaux" target="_blank">https://x.com/thsottiaux</a></p><p>• LinkedIn: <a href="https://www.linkedin.com/in/thibault-sottiaux-27195366" target="_blank">https://www.linkedin.com/in/thibault-sottiaux-27195366</a></p><p>—</p><p><strong>Where to find Lenny:</strong></p><p>• Newsletter: <a href="https://www.lennysnewsletter.com" target="_blank">https://www.lennysnewsletter.com</a></p><p>• X: <a href="https://twitter.com/lennysan" target="_blank">https://twitter.com/lennysan</a></p><p>• LinkedIn: <a href="https://www.linkedin.com/in/lennyrachitsky/" target="_blank">https://www.linkedin.com/in/lennyrachitsky/</a></p><p>—</p><p><strong>In this episode, we cover:</strong></p><p>(00:00) Introduction to Tibo Sottiaux</p><p>(01:12) How Tibo uses AI in his daily work</p><p>(03:10) The vision for always-on AI agents</p><p>(06:15) How AI will continue to change work</p><p>(07:32) Bringing Codex, ChatGPT, and Dots together</p><p>(08:20) The sci-fi inspiration behind the technology</p><p>(09:02) Building AI with the community</p><p>(10:13) Understanding the Dots structure</p><p>(11:34) OpenAI’s open ecosystem and plugin revenue sharing</p><p>(13:39) How plugins get discovered and recommended</p><p>(14:39) The research behind Dots</p><p>(16:20) Where humans will remain valuable</p><p>(17:25) Agent fatigue, loneliness, and context switching</p><p>(18:55) The pressure to do more with AI</p><p>(20:09) An agent that spotted a production outage before a live demo</p><p>(21:42) Agent guardrails and connecting multiple devices</p><p>(23:25) The skills becoming more valuable in the AI era</p><p>(25:56) Advice for people early in their careers</p><p>(27:18) How teams build and ship inside OpenAI</p><p>(28:59) Autonomy, trust, and learning from mistakes</p><p>(30:45) Building for a future where agents dominate internet activity</p><p>(32:41) What Tibo has changed his mind about</p><p>(33:55) OpenAI’s approach to AI safety and alignment</p><p>(36:15) Moving beyond model pickers toward simpler AI</p><p>—</p><p><strong>Referenced:</strong></p><p>• OpenAI Dots: <a href="https://chatgpt.com/features/dots" target="_blank">https://chatgpt.com/features/dots</a></p><p>• <em>Her </em>on Netflix: <a href="https://www.netflix.com/title/70278933" target="_blank">https://www.netflix.com/title/70278933</a></p><p>• <em>Star Trek</em>: <a href="https://www.imdb.com/title/tt0060028" target="_blank">https://www.imdb.com/title/tt0060028</a></p><p>• Pi: <a href="https://pi.ai" target="_blank">https://pi.ai</a></p><p>• OpenCode: <a href="https://opencode.ai" target="_blank">https://opencode.ai</a></p><p>• Notion: <a href="https://www.notion.com" target="_blank">https://www.notion.com</a></p><p>• Figma: <a href="https://www.figma.com" target="_blank">https://www.figma.com</a></p><p>• OpenClaw: <a href="https://openclaw.ai" target="_blank">https://openclaw.ai</a></p><p>• Peter Steinberger on X: <a href="https://x.com/steipete" target="_blank">https://x.com/steipete</a></p><p>• Grok Bot: <a href="https://x.ai/bot" target="_blank">https://x.ai/bot</a></p><p>• Muse: <a href="https://muse.ai" target="_blank">https://muse.ai</a></p><p>• Instinct: <a href="https://instinct.com" target="_blank">https://instinct.com</a></p><p>• Ahmed Ibrahim on LinkedIn: <a href="http://linkedin.com/in/ahmedibrhm" target="_blank">linkedin.com/in/ahmedibrhm</a></p><p>• The Hugging Face incident and the road ahead: <a href="https://openai.com/index/hugging-face-incident-and-the-road-ahead" target="_blank">https://openai.com/index/hugging-face-incident-and-the-road-ahead</a></p><p>—</p><p><strong>Recommended book:</strong></p><p>• <em>Neuromancer</em>: <a href="https://www.amazon.com/Neuromancer-William-Gibson/dp/0441007465" target="_blank">https://www.amazon.com/Neuromancer-William-Gibson/dp/0441007465</a></p><p>—</p><p>Production and marketing by <a href="https://penname.co/" target="_blank">https://penname.co/</a>. For inquiries about sponsoring the podcast, email <a href="mailto:[email protected]" target="_blank">[email protected]</a>.</p><p>—</p><p><em>Lenny may be an investor in the companies discussed.</em></p> <br /><br />To hear more, visit <a href="https://www.lennysnewsletter.com?utm_medium=podcast&utm_campaign=show-notes-no-free-preview-language">www.lennysnewsletter.com</a>
Key Insights
- Sottiaux experiences cycles of expanding agent teams when approaching capability limits, then consolidating them when model breakthroughs occur, suggesting a dynamic equilibrium in how AI complexity is managed.
- The majority of future internet actions will be taken by agents, requiring products to be architected for agent-scale usage with associated strain on infrastructure and economics.
- OpenAI deliberately chose not to release a more capable model (6.1 Astra) despite having developed it, demonstrating a deliberate pacing strategy informed by safety and alignment considerations.
- Voice interaction has become more critical to Sottiaux's workflow than anticipated, allowing return to creative flow states that were previously achieved through traditional engineering work.
- The interview reveals that 'loops' and manual configuration in AI systems are being replaced by self-learning systems that adapt based on feedback, representing a fundamental architectural shift.
- OpenAI's safety investment now rivals or exceeds primary model development investment, with parallel compute dedicated to monitoring agents for prompt injection and high-risk actions.
- New workforce entrants have a competitive advantage over experienced workers because they lack ingrained habits from pre-AI workflows and can adopt AI-first approaches more readily.
- Building delightful experiences for humans that leverage multimodal AI capabilities is described as under-invested compared to technical capability development.
Topics
Transcript
Often when I look at what people are building out there, it's just like, you're not quite getting it. You know, if you were just like pushing yourself and just really imagining like all of this being roughly 10 times better than it is today, you know, like in a year, you would build in a different way. What do you think people aren't pricing in in terms of where things are going? The majority of actions on the internet will be taken by agents. If you want your product to be successful for agents, you have to build, you know, for a certain level of scale. How many agents do you find yourself running in parallel just uh in…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Lenny's Podcast: Product | Career | Growth
The grief, loneliness, and burnout sweeping through the tech industry right now | Molly Graham
Molly Graham revisits her famous 'give away your Legos' career advice in light of AI disruption, finding that while the core message about embracing change remains valid, the advice requires significant caveats when applied to delegating work to AI rather than humans. She argues that some work shouldn't be outsourced to AI, emphasizes the importance of acknowledging grief and emotions around change, and reframes AI as an intern rather than a superior intelligence.
90 minutes of unfiltered product advice from Snap and Discord’s product chief | Peter Sellis
Peter Sellis, former head of product at Snap and Discord, discusses building high-performing product organizations by organizing teams like ideological structures with clear decision-making authority, argues that most product managers are mathematically bad due to power-law distributions in skill, and shares lessons on focusing growth on core product value rather than expansion.
How we built Grok Bot in a month | Roman Ugarte (SpaceXAI)
Roman Ugarte, product lead at Grokbot (SpaceX AI), discusses how they built and launched a groundbreaking AI agent product in just one month by making two critical early decisions: running bots in the cloud with their own computers, and focusing ruthlessly on simplicity over features. The conversation covers their go-to-market strategy, the importance of manual onboarding, and how maintaining startup culture enables rapid iteration in a highly competitive AI market.
Why companies are becoming a series of loops | Anish Acharya (a16z)
Anish Acharya, a16z General Partner and former founder, discusses how AI is reshaping company organization through 'loops' (autonomous agent workflows), argues the permanent underclass fear is overblown, and contends that the real consumer opportunity lies in AI applications that improve happiness and human connection rather than just productivity.
AI’s third era: the rise of persistent AI coworkers | Tara Seshan (OpenAI’s product lead)
Tara Seshan, OpenAI's product lead for ChatGPT and Codex, discusses how AI is fundamentally changing product management and knowledge work. She emphasizes the shift from theoretical planning to rapid empirical iteration, the importance of maintaining human ambition and opinionation in an age of AI capabilities, and how products like work mode and sites are enabling persistent AI coworkers.