Technology
OpenAI Presence Moves Enterprise Agents From Pilots Into Production Workflows
OpenAI introduced Presence, an enterprise product for deploying voice and chat agents with policies, guardrails, approved actions, evaluations, and human escalation built into customer workflows.
By Patrick T ยท
OpenAI has introduced Presence, a new enterprise product that shows how the AI agent market is moving away from demos and toward controlled production work. The company said on July 22 that Presence is available to eligible enterprise customers through a limited general availability program and is designed for voice and chat agents that can answer questions, resolve issues, use company systems, take approved actions, and escalate to people when needed. That framing matters because the hard part of enterprise AI is no longer proving that a model can hold a conversation. It is proving that an agent can operate inside a business without turning permissions, data access, and accountability into an afterthought.
Presence starts with a specific job rather than a broad assistant promise. OpenAI listed examples such as billing issues, insurance claims, and employee IT service requests. In each case, the agent is supposed to receive only the knowledge and system access required for that workflow. The company sets policies for what the agent can do, when approval is required, and when a human should take over. That is a sober product design choice. Enterprise customers do not need a free-roaming chatbot. They need a bounded worker that knows the local policy, can use the right tools, and can be stopped before a bad answer becomes an operational decision.
OpenAI says the product combines policies and standard operating procedures, guardrails, approved actions, simulations, evaluation tools, and a Codex-powered improvement process. That combination is the real product. The model is only one component. The surrounding system decides what data is available, what action is legal, how performance is graded, and how a proposed change reaches production. Presence therefore reads less like a chatbot launch and more like an attempt to package the control plane that many companies have been building manually around AI tools.
The early numbers OpenAI disclosed are specific but still company-reported. Presence powers OpenAI's English-language phone support channel at 1-888-GPT-0090, and the company said it now resolves 75% of inbound issues without human assistance. OpenAI also said its Codex-powered improvement loop reduced human handoffs by 15 percentage points in 10 days. Those claims are useful because they describe operating metrics, not only model benchmarks. They still need to be read as early vendor evidence until customers publish more independent performance data.
Business Insider reported that Presence is part of OpenAI's shift beyond the model war and into corporate software. That is the right way to understand it. The frontier model market is becoming more competitive and more price-sensitive. Enterprises can increasingly choose among OpenAI, Anthropic, Google, Meta, and open models for different tasks. If model access alone becomes easier to substitute, the durable value moves into deployment expertise, workflow integration, and the evidence that a system keeps working after launch.
The customer list points to the kinds of workflows OpenAI wants to own. BBVA is exploring AI-powered voice support for banking customers in Mexico. SoftBank is testing natural Japanese-language customer conversations. IAG is exploring support during severe weather and other high-demand insurance moments. Those are not low-risk toy settings. Banking, telecom, and insurance all involve identity, sensitive records, policy exceptions, and regulatory exposure. A useful agent must know when not to act as much as when to act.
The product also clarifies the role of OpenAI's forward-deployed engineers. Presence is not self-serve. OpenAI says deployments are led by its own Forward Deployed Engineers and selected global systems integrators. That means the company is selling an implementation relationship as well as software. The advantage is that difficult workflows can be shaped with the vendor close to the customer. The risk is that customers may become dependent on OpenAI's way of encoding business process, evaluation, and change management.
For enterprise technology leaders, Presence should be evaluated on narrower questions than the launch language suggests. Which systems can the agent access? Which actions require approval? How are failed sessions reviewed? Can policies be audited? Can the customer export logs and evaluation data? Can another model provider be substituted later? Those questions determine whether an agent platform becomes critical infrastructure or another specialized support tool.
The broader market signal is that AI agents are being professionalized. The industry spent two years talking about autonomy. Presence is a bet that autonomy will be bought only when it comes with boundaries. The companies that deploy agents will still need their own security reviews, legal checks, and workflow owners. OpenAI can package more of that work, but it cannot remove the customer's accountability for what the agent does.
That is why the launch matters. Presence makes OpenAI more like an enterprise software vendor, a services partner, and a model lab at the same time. If it works, the next phase of AI adoption will be judged less by how fluent agents sound and more by whether they can close tickets, follow policy, document decisions, and hand the work back to people when judgment is required.
Topics: OpenAI, Presence, enterprise AI, AI agents