OpenAI has spent two years telling enterprises that agents are ready for production. This week it started selling the part that was actually missing: the operational scaffolding around them. OpenAI Presence is a new enterprise product for deploying voice and chat agents that answer questions, resolve issues, act inside company systems, and — critically — hand off to a human when the rules say so.
The pitch is blunt about where enterprise AI actually stalls. In OpenAI’s own framing, the challenge “is no longer proving that AI agents can work, it’s making them reliable enough to do high-value work in production.” Presence answers that with policies, guardrails, approved actions, simulations, evaluation tooling, and an improvement loop driven by Codex, OpenAI’s coding agent.
A product shaped like a deployment, not a model
Presence deployments don’t start with a model picker. They start with a job: resolving billing issues, supporting insurance claims, handling employee IT requests. The agent receives only the knowledge and system access that specific job requires, and the company defines what it may do on its own, what needs approval, and when a person takes over.
The more unusual piece is what happens after launch. Production sessions and escalations surface the gaps — a policy the agent misread, a customer behavior nobody simulated. Codex, running with a Presence plugin, investigates those signals and proposes updates. Teams test each proposed change against the version already in production, then approve a controlled rollout. It is continuous improvement with a paper trail, which is exactly the shape of change management enterprise compliance teams already understand.
Before anything reaches users, teams can run the agent through simulations covering common requests, edge cases, and higher-risk scenarios, with graders checking whether it reached the right outcome, followed policy, used tools correctly, and escalated when appropriate.
OpenAI is its own first case study
The strongest evidence in the announcement is internal. Presence powers OpenAI’s English-language phone support line at 1-888-GPT-0090, where it verifies callers, uses account context, and takes approved actions. OpenAI says the system now resolves 75% of inbound issues without human assistance, and that the Codex-powered improvement loop cut human handoffs by 15 percentage points in just 10 days after launch.
Named early adopters are exploring narrower slices: BBVA is looking at AI-powered voice support for everyday banking in Mexico, SoftBank is testing natural Japanese-language customer conversations, and IAG is exploring surge support during high-demand events like severe weather.
Independent coverage picked up the competitive subtext quickly. AI News framed the launch as enterprise agents “with engineers included,” while contact-center trade press read it as a direct move into CX territory long held by specialist vendors — No Jitter headlined it as OpenAI making “its Presence felt in CX.”
The catch: you can’t just sign up
Presence is not self-serve, and OpenAI is unusually explicit about that. It’s available to eligible enterprise customers through a limited general availability program, with deployments led by OpenAI’s Forward Deployed Engineers and select global systems integrators. When a use case goes beyond what the product supports, those FDEs and partners work with the customer to bring it into production anyway.
That staffing model is the real story. OpenAI isn’t just shipping software; it’s selling deployed outcomes with its own engineers attached — the business model of Palantir and the big consultancies, not of an API vendor. Every deployment also feeds back: OpenAI says generalized insights from customer deployments inform its research and product development, compounding across the customer base.
For teams already building voice agents on the raw API, nothing is being taken away — OpenAI says it will keep supporting voice customers with frontier-model access through the API. But the ladder now has a clear top rung, and it comes with an account team.
What to watch
The open questions are pricing, which OpenAI hasn’t published, and how far the limited availability program widens. If the 75% self-serve resolution rate holds outside OpenAI’s own support line — in a bank’s billing queue or an airline’s weather-delay surge — Presence stops being a product announcement and becomes a benchmark every contact-center vendor has to answer. Enterprises evaluating it today should ask the sharper question: not whether the agent can do the job, but whether the guardrail, escalation, and change-approval model fits how their compliance team already works. That, not the model, is what OpenAI is actually selling.
