OpenAI Presence turns chatbots into supervised digital workers

OpenAI has introduced Presence, a managed enterprise product designed to help companies deploy AI agents for customer service and internal operations. The system supports real time voice and chat interactions, allowing agents to answer questions, access selected company systems, complete approved tasks and hand difficult cases to employees.

Presence is not a self service tool that businesses can configure independently. OpenAI says deployments are led by its Forward Deployed Engineers, supported in some cases by systems integration partners. The product is currently available only to eligible enterprise customers through a limited general availability program.

The company presents Presence as an answer to a common problem in enterprise AI: moving from impressive demonstrations to dependable daily operations. An agent may work well when it is first launched, but company policies, products and customer requests change over time. Presence is intended to give businesses a structured way to monitor those changes and update an agent without allowing it to alter its own behavior unchecked.

Controls around the AI agent

Each Presence deployment begins with a narrowly defined task. OpenAI cites examples such as handling billing disputes, supporting insurance claims and processing internal IT service requests. Companies decide which knowledge sources and software systems the agent can access. They also set boundaries for actions that the agent may complete on its own, actions requiring approval and situations that must be transferred to a person.

The product combines several operational tools:

  • Company policies and standard operating procedures
  • Guardrails that limit behavior outside defined rules
  • Approved actions, such as processing an eligible refund
  • Simulations and evaluations before launch
  • Escalation rules for human intervention
  • Monitoring of production conversations and outcomes

Before an agent goes live, teams can test common customer questions, unusual cases and higher risk scenarios. Automated graders assess whether the agent reached the intended result, followed policy, used tools correctly and escalated the conversation when required.

After launch, Presence collects signals from conversations, failed interactions and human handoffs. OpenAI says Codex, using a Presence plugin, can examine those signals and suggest changes. A company team must test and approve each proposed update before it is released.

OpenAI reports that Presence powers its English language phone support line, where it handles open ended requests, verifies callers and uses account information to complete approved actions. The company says the agent resolves 75 percent of inbound issues without human assistance and that its improvement process reduced handoffs by 15 percentage points in ten days. These figures are company reported and do not define measures such as repeat contacts or customer satisfaction.

BBVA Mexico is working with OpenAI as a design partner for voice based financial service. VentureBeat reports that SoftBank is testing Japanese language customer conversations, while Insurance Australia Group is exploring support during severe weather and natural disasters. None of those examples has been described as a full production rollout.

Important questions for buyers

Presence places OpenAI closer to the contact center software market, where companies such as Salesforce, Genesys, NiCE, ServiceNow and Amazon offer AI assisted service tools. However, the available information does not indicate that Presence replaces a complete contact center platform. Features such as workforce management, broad interaction routing and established reporting may still require other systems.

OpenAI has not disclosed pricing, supported integrations, service commitments or the specific models used in Presence. For companies considering the product, the most important questions will concern permissions, audit records, data handling, incident response and the measurement of customer outcomes. The value of a customer service agent will depend not only on how many conversations it resolves, but also on whether it gives correct answers, protects data and knows when to involve a human.

Sources

Stay up to date

AI for content creation: the latest tools, tips and trends. Every two weeks in your inbox:

More info …

About the author

Related posts:

Advertisement

×