OpenAI launches Presence for enterprise production voice and chat agents

OpenAI announced Presence today, an enterprise platform for deploying and managing production AI agents across voice and chat channels. The product is available immediately through a limited general availability program, with deployments led by OpenAI's Forward Deployed Engineers and select authorized systems integrators — not available on a self-service basis.

The core problem Presence targets is not whether AI agents can work, but whether they can stay reliable as business conditions change. Policies shift, products get updated, and customer behavior evolves — and an agent that performs well at launch can degrade months later without a formal mechanism for detecting and correcting that drift. Presence packages the policies, guardrails, simulation tooling, and a Codex-powered update loop designed to maintain agent reliability after launch.

How it works

Each Presence deployment defines a specific job — resolving billing disputes, processing insurance claims, handling IT service requests — and scopes the agent's knowledge and system access to exactly that task. The enterprise sets the policies: what the agent can do independently, which actions require approval, and when a human must take over.

Before production, teams test the agent against common requests, edge cases, and high-risk scenarios using simulation batches. Graders check whether the agent reached the intended outcome, followed the defined policy, used tools correctly, and escalated when required. After launch, production sessions and escalation patterns surface where the agent is working and where it is not.

Codex then investigates those gaps and proposes behavioral updates. Teams review and approve changes before they go live, and a controlled rollout compares the proposed update against the existing production version before full deployment. The loop runs continuously, intended to keep agent behavior aligned with current policies without giving an automated system authority to rewrite itself unchecked.

Proven in production

OpenAI is using Presence to run its own English-language phone support at 1-888-GPT-0090. According to the company, the agent meets or exceeds the benchmarks used to grade human frontline support quality and resolves 75% of inbound issues without any human handoff. The Codex improvement loop reduced human handoffs by 15 percentage points in the first 10 days after launch.

Enterprise early adopters include BBVA, which is exploring AI voice support for everyday banking in Mexico; SoftBank, which is testing natural Japanese-language customer service conversations; and IAG, the parent company of British Airways and Iberia, which is exploring time-sensitive support during high-demand events such as severe weather disruption.

What is not yet available

OpenAI has not disclosed pricing for Presence, nor the cost of the engineering and integration work that accompanies each deployment. The product launches with voice and chat support; email and additional channels are mentioned as a future direction but are not confirmed at launch. Presence uses OpenAI models for the core agent layer, while allowing customers to connect third-party models through APIs for guardrails, tools, and other workflow components.

The non-self-service delivery model — where every deployment requires OpenAI engineers or certified partners — positions Presence as a managed service rather than a developer platform, placing it in direct competition with enterprise AI consulting offerings and agentic platforms like Salesforce Agentforce, as first reported by VentureBeat.