The enterprise AI race is moving from impressive demonstrations to managed operations.
OpenAI has launched Presence, a deployed product for building and operating voice and chat agents across customer service and internal workflows. These agents can answer questions, access company knowledge and systems, take permitted actions, and escalate cases to people when judgment or approval is required.
The important part is not simply that the agent can act. Presence packages that capability with the controls needed to decide what it may access, which actions it may take, how it is tested, and how its behavior changes after launch.
From a chatbot to an operating role
A conventional chatbot is usually placed in front of a knowledge base. A Presence deployment is scoped around a business workflow. OpenAI works with the customer to identify the job, connect the required systems, establish permissions and policies, test the agent, and bring it into production.
That could mean resolving a customer request, helping an employee navigate an internal process, updating a record, or carrying a case through several systems. The agent receives the information and access needed for that role rather than broad, undefined authority.
Companies determine when the agent may act independently, when approval is required, and when the work must be handed to a person. Human escalation is therefore part of the workflow design, not an emergency feature added after deployment.
Testing becomes part of the product
Before an agent goes live, Presence uses simulations and automated graders to test its behavior. The goal is to see whether it follows policy, handles edge cases, completes the intended task, and recognizes situations that should be escalated.
This matters because a production agent does not operate on a fixed set of prompts. Customers ask unusual questions, company policies change, connected systems return imperfect data, and actions can have consequences beyond the conversation itself.
A useful evaluation program therefore needs representative scenarios, explicit success criteria, policy checks, adversarial cases, and a record of how changes affect performance. The agent is not merely launched; it is measured against the job it is expected to do.
Operations continue after launch
Presence is designed around continuous operation rather than one-time deployment. Teams can monitor real interactions, identify failures and new patterns, evaluate proposed improvements, and approve controlled updates.
That creates a safer improvement loop. A new instruction, tool connection, policy, or workflow change can be tested before it alters live behavior. Organizations can compare results and decide whether an update is ready instead of allowing an agent to drift through informal prompt changes.
The model is only one component. The operational system also includes identity, permissions, connected data, tool execution, policy enforcement, evaluation, monitoring, incident handling, and ownership.
Why it matters
Enterprise agents have spent years looking capable in controlled demonstrations. Production is harder. Real workflows contain exceptions, conflicting objectives, private data, approvals, legacy systems, frustrated users, and actions that may be difficult to reverse.
Presence is OpenAI's attempt to package the less glamorous infrastructure businesses need before allowing an agent to perform consequential work. It shifts the product question from “Can the model complete this task?” to “Can the organization operate this agent reliably over time?”
That is a more mature standard. Capability still matters, but adoption will increasingly depend on whether the entire system is observable, governable, and recoverable when something goes wrong.
The SunMarc angle
For SunMarc App Labs, the lesson is that trustworthy automation is a product system, not a model feature. Even a small agentic workflow benefits from explicit permissions, narrow tools, test cases, approval gates, visible logs, and a clear path back to a person.
Those controls are not only for large enterprises. A support assistant, content workflow, data utility, or app feature should define what the automation owns, what it may change, and how a user recovers from an incorrect action.
The strongest products will make operational trust visible. Users should understand what the agent can do, why it did something, when a human is involved, and how to correct the result.
The product lesson
Do not measure an agent only by the percentage of tasks it can complete. Measure the percentage it completes correctly within policy, the quality of its escalations, the reversibility of its actions, and the cost of detecting and correcting failure.
OpenAI Presence suggests that enterprise AI is becoming an operations discipline. The winning agent platforms will not merely produce better answers. They will give organizations a controlled way to deploy, observe, improve, and govern AI work.
Presence is currently available to eligible enterprise customers through a limited general availability program. It is a managed deployment rather than a self-service agent builder inside a ChatGPT workspace.
Relevant links
- OpenAI: Introducing OpenAI Presence
- OpenAI Help Center: OpenAI Presence
- TechRadar: OpenAI launches its enterprise agent product
- SiliconANGLE: OpenAI Presence targets enterprise agents
- SunMarc archive: GPT-5.6 Turns OpenAI's Model Launch Into a Work Platform
- SunMarc archive: Enterprise AI Buyers Are Starting to Push Back on the Labs