Personal AI Agents, Product Roasts, and the Dots Bet

Three different companies are converging on the same hard problem this week (how much access and autonomy a personal AI agent should get) while Notion quietly ships the infrastructure layer (agent skills, shared memory) that makes any of this usable at scale. ### OpenAI / Lenny Rachitsky - Why Dots Is OpenAI's Biggest Bet, According to Its Own Head of ChatGPT **Source:** https://www.lennysnewsletter.com/p/openais-head-of-chatgpt-were-entering **Credibility:** High (direct interview with Tibo Sottiaux, OpenAI's Head of ChatGPT, on Lenny's podcast) **What happened:** Lenny interviewed Tibo Sottiaux, who runs ChatGPT at OpenAI, about why Dots (OpenAI's agentic product, launched at DevDay) is the company's biggest internal bet right now. Sottiaux's argument: agents are going to account for a growing share of internet traffic, and the builders who are still treating this like "chatbot plus plugins" are missing where the puck is going. **Key capabilities:** - Sottiaux frames Dots as a bet on agents replacing a meaningful share of how people interact with the web, not just a chat feature bolt-on - The interview digs into what builders are "still getting wrong" when designing for agent-first usage, specifically around handing off tasks vs. keeping humans in a chat loop - This comes right after OpenAI's DevDay, where Dots was one of 20+ launches but positioned internally as the headline bet, per Sottiaux **Why it matters for PMs:** This is a rare moment where you get the internal rationale straight from the person running the product, not a press release. If OpenAI's own leadership is explicitly calling Dots (not GPT-6, not Sol/Luna) the biggest bet, that's a signal about where OpenAI thinks the margin and distribution fight actually is: agentic task completion, not model quality. For anyone building a product that touches ChatGPT (plugins, GPTs, API integrations), this is a signal to re-evaluate whether your integration still works in an agent-first world, or whether it was built for a chat-first one that's already fading. **Critical questions:** - "Agents will dominate internet traffic" is a bold claim. What's the actual usage data behind Dots today versus this being aspirational framing? - If Dots succeeds, what happens to the long tail of GPTs and plugin-based integrations built for the old paradigm? - Sottiaux is the product leader selling the bet internally and externally. What's the view from engineering or research on whether Dots is actually ready for this scale of traffic shift? **Action you could take today:** If your product has any ChatGPT integration (plugin, Actions, API), spend 30 minutes auditing whether it's designed for a user typing a prompt or an agent executing a multi-step task autonomously. Those are different integration surfaces.