OpenAI Head of ChatGPT and Codex Tibo Sottiaux Predicts Model Pickers Will Disappear as Agents Take Over

Tibo Sottiaux, Head of ChatGPT and Codex at OpenAI, asserts that model pickers are a temporary friction point that will disappear as AI systems become more autonomous. He states that Dots ships with no model picker at all, allowing users to simply talk to the agent, and predicts that most actions on the internet will soon be taken by agents rather than humans. Sottiaux notes that when Notion launched its MCP server, the resulting flood of agent traffic strained systems and forced a rethink of economics, a scenario he believes is inevitable for every product team.

Building for Future Capabilities

Sottiaux advises developers to build as if models will be 10 times better, cheaper, and faster within a year. He criticizes the current reliance on complex agent loops, graphs, and fine-tuned prompts as a passing phase, arguing that future models will handle these tasks seamlessly. His own forecast for current capabilities was too slow, expecting them to arrive a year or two from now, so he urges teams to anticipate rapid improvement rather than optimizing for today's limitations.

Agent Dynamics and Hiring

Sottiaux describes a cyclical approach to agent teams: he expands them when pushing the frontier, then shrinks them when a new, more capable model lands that can handle the workload with better memory and learning. He emphasizes that taste and user understanding are becoming more valuable than technical speed, noting that OpenAI employs more than 120 former founders. He also highlights a shift in hiring strategy toward new graduates, citing Ahmed Ibrahim, who built much of the Codex harness and now manages OpenAI’s compute fleet, as an example of young talent rapidly mastering these tools.

Safety and Burnout

In a practical demonstration of agent autonomy, Sottiaux’s Dot agent warned him of a production outage five minutes before a live DevDay demo, correctly deducing the dependency and offering to fix it. Looking forward, Sottiaux argues that AI should reduce noise and burnout rather than increasing pressure to run multiple agents simultaneously. He envisions a future where AI allows for better rest and creativity, stating the goal is not just "one more prompt per second."