The VP who built OpenAI's new assistant just revealed he won't let his AI talk to yours, and that design choice tells you everything about where agents are actually headed.

The Summary

  • OpenAI launched dots, an always-on personal assistant evolved from its Codex coding tools, but its creator Alexander Embiricos refuses to let it become another scheduling bot
  • The product philosophy: agents should amplify human decision-making, not replace human interaction with bot-to-bot negotiations
  • This is OpenAI's first major consumer product pivot since ChatGPT, signaling a shift from conversation to continuous ambient assistance

The Signal

OpenAI just made its coding infrastructure personal. Dots emerged from the same foundation as Codex, the system that powered GitHub Copilot and generated code from natural language. Now it's running in the background of your daily life, watching context, suggesting actions, handling tasks you'd normally delegate to a junior employee.

The technical lineage matters. Coding assistants succeeded because they understood structured environments with clear right answers. Dots is betting that personal productivity has more structure than we think: email patterns, calendar logic, document workflows, research habits. These aren't creative acts. They're algorithmic drudgery dressed up as knowledge work.

"I never want to have a meeting with someone else's AI agent." — Alexander Embiricos

But Embiricos drew a hard line that most agent companies are ignoring. He doesn't want dots negotiating meeting times with your assistant while you both sit out of the loop. He doesn't want the AI-to-AI economy that venture capital has been funding for two years. This isn't a technical constraint. It's a product philosophy about where agency actually lives.

Key tensions in the always-on assistant model:

  • Ambient monitoring requires trust that most consumer AI hasn't earned yet
  • Continuous access to email, calendar, and documents creates new security surfaces
  • The line between "assistant" and "replacement" gets blurrier when the human isn't actively prompting

The real tell is what OpenAI chose NOT to build. Every competitor is racing toward autonomous agents that complete multi-step tasks without human input. Dots pulls back from that. It watches, it suggests, it executes when asked. The human stays in the loop. That's either the right bet on how people actually want to work with AI, or it's a temporary hedge before OpenAI releases the full autonomous version.

This also marks OpenAI's first serious attempt at a persistent consumer product beyond ChatGPT. Chat is ephemeral by design. Dots is persistent. It's learning your patterns over weeks and months, not just responding to a single prompt. That shift from session-based to continuous is the actual product innovation here, more than any specific feature.

The Implication

Watch how enterprise buyers respond to the ambient monitoring model. If dots succeeds, it proves that people value AI amplification over AI automation, at least for knowledge work. If it fails, expect OpenAI to quietly walk back the human-in-the-loop philosophy and ship the autonomous version their research team probably already built.

For anyone building agents: the Embiricos stance is your fork in the road. Do you build for a world where AIs negotiate with each other, or where they make humans more effective? Both paths are viable. Only one aligns with how OpenAI is currently betting.

Sources

TechCrunch AI