Lenny Rachitsky (@lennysan) on X
Short summary
Tibo Sottiaux, head of ChatGPT and Codex at OpenAI, predicts that agents will handle most internet actions soon, the model picker will disappear, and people should build assuming models will be 10x better in a year. He also argues that complex agent architectures are a passing phase and that taste—understanding users and knowing what good looks like—will matter more than technical skills like typing speed.
- Agents will soon take most actions on the internet, forcing every product team to decide whether to build for them; Notion's MCP server launch caused system strain and forced economic rethinking.
- Models will improve at "incredible" rates in speed and cost over the next year, so builders should design as if today's capabilities are 10x worse than what's coming, not optimize for the present.
- Complex agent loops and graph architectures are a temporary solution—as models get better, larger single agents can handle what previously required teams, making intricate setups obsolete.
- The OpenAI plugin ecosystem could be a major opportunity: plugins get revenue share when users spend tokens, and ChatGPT recommends them to 1.2 billion users based on retention and quality.
Full summary
Tibo Sottiaux, head of ChatGPT and Codex at OpenAI, thinks the AI landscape is moving faster than most people realize. Agents will soon handle most internet actions—Notion's MCP server launch caused traffic spikes that forced the company to rethink its economics, a preview of what every product team will face. The friction points users deal with now, like choosing between models, will disappear. Sottiaux built Dots with no model picker at all because he finds the current setup exhausting. His bigger bet: build as if models will be 10x better in a year. He expects speed and cost improvements at rates he calls "quite incredible," with all modalities working seamlessly together. When he looks at what people are building today, his instinct is that they're not thinking ambitiously enough.
The technical approaches getting hype right now won't stick around. Intricate agent loops and graph architectures had their moment, but Sottiaux sees that era ending. Instead, as models improve, he's noticed his own agent teams expand and contract with each release—larger teams when pushing boundaries, then shrinking back when a single more capable agent can handle everything. The plugin ecosystem might matter more than people think. OpenAI launched with 16 partners for "Sign in with ChatGPT," and popular third-party plugins will share revenue when subscribers spend tokens. The platform recommends plugins based on retention and quality, potentially exposing good ones to 1.2 billion users. His advice to developers is straightforward: build something good.
The skills economy is shifting. Typing fast is losing value fast; taste—understanding users deeply, learning quickly, knowing what good looks like—is what matters now. That's why OpenAI has over 120 former founders on staff. Sottiaux also changed his mind about hiring new graduates. He underestimated how quickly younger people would embrace AI tools and learn at scale. Ahmed Ibrahim, hired as a new grad, built much of the Codex harness and now runs OpenAI's compute fleet. What set him apart wasn't credentials—it was being kind, collaborative, and willing to absorb new tools at speeds Sottiaux hadn't seen. The final piece is burnout. There's pressure to run 30 agents at once and ship constantly because you can. Sottiaux wants AI to reduce noise instead, letting people focus where it matters and actually rest. The promise isn't just more prompts per second; it's better work from people who aren't burned out.