← Back to feed

OpenAI launched Presence on August 6, a productised enterprise deployment offering that bundles the company’s years of in-house deployment expertise into a repeatable managed service. The positioning is explicit: enterprises no longer need to prove AI agents can work, they need to make them reliable enough to handle high-value work in production.

What it is

Presence is not an API product. It is a deployment engagement where OpenAI — and select systems integrators — work alongside enterprise customers to define, test, and operate a production AI agent.

Each deployment starts with a single scoped job: resolving billing disputes, handling insurance claims, managing employee IT service requests. The agent receives only the knowledge and system access required for that specific workflow. The company defines the policies: what the agent can and cannot do, when it needs human approval, when it escalates entirely.

After launch, the deployment enters a Codex-powered improvement loop. Production sessions and escalations reveal behavioural gaps, Codex proposes updates, and customer teams approve or reject them before they deploy. The loop is designed to let agents adapt as products, policies, and user behaviour change — without requiring the enterprise to rebuild from scratch each time.

The components

Presence brings five elements together in one offering:

• Policies and SOPs: Codified behavioural rules for the agent
• Guardrails: Constraints on what the agent can and cannot do
• Approved actions: A defined set of permitted system calls and operations
• Simulations: Pre-launch testing against expected scenarios
• Evaluation tooling: Ongoing performance and accuracy monitoring

Today, Presence supports voice and chat channels — customer support, outbound sales, and what OpenAI describes as “high-risk internal workflows.”

Why this matters

OpenAI is entering territory that has historically belonged to system integrators and large consulting practices. Accenture’s AI studios, Deloitte’s AI implementation arm, and the growing class of AI implementation boutiques all sell some version of what Presence describes. The difference is OpenAI’s claimed compounding advantage: every enterprise deployment generates production data that feeds back into OpenAI’s research, and those insights propagate to all Presence customers.

That claim is either the most valuable thing about Presence or the thing that will make general counsels very nervous.

The other notable element is the explicit consulting model. Presence is not self-serve. OpenAI assigns staff to each deployment, and the offering expands via named systems integrators. For a company that spent years selling API credits, this is a significant motion toward professional services.

Competitive context

Salesforce AgentForce, ServiceNow AI Agents, and Microsoft’s Copilot Studio all occupy adjacent positioning at the enterprise layer. Each is a platform product — buy once, deploy many. Presence is pitched as something closer to a managed outcome: OpenAI stays in the loop, Codex keeps iterating, and the agent improves without requiring the customer to manage a prompt engineering team.

Whether enterprises pay for that model depends on whether the managed approach produces measurably better agents than self-serve alternatives. OpenAI has not published comparative performance data.

Pricing has not been disclosed.