TiboSottiaux

Tibo Sottiaux, a longtime engineer now leading a major part of OpenAI's product and Codex work, discusses how agents, persistent assistants like dots, and plugin ecosystems are reshaping how people build and work. The conversation covers the future of AI interfaces, the shifting skills that matter for hiring, OpenAI's internal culture, and safety tradeoffs at the frontier.

01Key learnings

• Expect most actions on the internet to be taken by agents, so build products that can serve agent traffic at scale and handle the economics that come with it.
• Design for the capabilities you expect in roughly a year (about 10x better than today), not just what current models can do, to avoid building the wrong thing.
• Favor systems that learn from your goals and feedback over hand-tuned agent loops and workflows, which Tibo thinks will be short-lived as models improve.
• When models breakthrough, shrink agent teams: a bigger model can often do the work of many smaller agents, so keep reassessing team structure.
• For plugin ecosystems, success is measured by sustained retention and real utility; plugins that users keep using get recommended more.
• Hiring signals are shifting: typing speed matters less, while taste, user empathy, and founder-style ownership matter more.
• Keep safety investment ahead of capability, including secondary monitoring compute that watches the primary agent for risky actions or prompt injection.
• Simplify the product surface as models improve; reduce configuration like model pickers so users can just talk to the system.

Summary and takeaways written by PM Atlas; quotes are short excerpts. © the original author.