OpenAI’s Alexander Embiricos on Dots, Persistent AI Assistants, and Enterprise Adoption

An interview with Alexander Embiricos, head of enterprise product at OpenAI, about the development and positioning of Dots, OpenAI’s new personal AI assistant. The discussion traces how chatbot interfaces evolved into persistent, action-oriented agents and examines the product decisions behind making agent capabilities accessible beyond technical users. It also addresses trust, privacy, permissions, and how agent ecosystems may interact with businesses and consumers.

■ Evolution of AI assistants and Dots
- From ChatGPT chat to agents that can act
- Persistent, always-on assistant design
- Anthropomorphism, naming, avatars, and single-conversation context

■ Product differentiation and model choices
- Dots built on GPT-6 Astra for capability and reliability
- Integration with user’s computer and connected services
- Comparison with other agent products such as Meta Muse, Instinct, and Hermis

■ Privacy, control, and permissions
- Activity logs, connector access, and conservative defaults
- Fine-grained read/write and file/folder permissions
- User-configurable instructions versus deterministic controls

■ Real-world use cases and limitations
- Proactive research, flight booking examples, Slack and docs workflows
- Context degradation concerns over longer use
- Enterprise and pro-user focus versus broader consumer rollout

■ Ecosystem and future outlook
- Agent-to-agent interactions and business integrations
- Delegation versus increased productivity
- Long-term vision for persistent assistants

Viewers interested in AI product strategy, agent design, and enterprise AI adoption will find an overview of OpenAI’s current approach and open questions. The interview provides context on Dots’ target users, permission model, and the trade-offs involved in building reliable assistants.