OpenAI Presence: agents you rent, not build

OpenAI has announced Presence, and it signals where the AI agent race is actually heading: not toward smarter chatbots you build yourself, but toward fully managed agents that big companies deploy, govern, and improve over time. Presence is an enterprise platform for putting production-grade AI agents on the phone and in chat — for customer support, sales, and internal operations.

The pitch isn't the model — it's everything around it. Presence bundles company policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tools, and a Codex-powered improvement loop, so an agent stays accurate and predictable even as products and policies change. In other words, OpenAI is selling the hard part of shipping an agent that won't embarrass you in front of customers: governance.

Why it matters

For most of the last two years, "AI agents" meant a demo you wired together and hoped held up. Presence reframes them as a managed product with controls, monitoring, and accountability — the things an enterprise legal and ops team actually require before an agent touches a customer. The proof point is OpenAI's own English-language phone line, where Presence already resolves 75% of inbound calls without a human; BBVA Mexico, SoftBank, and insurer IAG are among the first customers.

There's a catch worth noting: this isn't a self-service product. It launches as a limited general-availability program, deployed by OpenAI's own Forward Deployed Engineers and select integrators — you rent the capability and the expertise, you don't just sign up. That's a very different model from the tools most people use day to day, and a sign that the highest-value agent work is consolidating around a few providers who can operate it. For everyone else, the same shift shows up as agent features landing inside the tools you already use.