OpenAI Launches Presence for Enterprise Voice and Chat Agents
OpenAI launched Presence on July 22 as a limited-availability enterprise product for deploying voice and chat agents across customer and internal workflows. It combines scoped system access, policies, guardrails, simulations, evaluations and human-approved Codex suggestions; OpenAI says its own support deployment resolves 75% of inbound issues without human assistance, a company-reported result not independently benchmarked.
OpenAI launched Presence on July 22 as an enterprise product for deploying voice and chat agents in customer-facing and internal workflows. The product is available to eligible enterprise customers through a limited general-availability program led by OpenAI's Forward Deployed Engineers and selected systems integrators; it is not a self-service offering.
What Presence includes
Presence packages the operational controls around an agent rather than introducing a standalone model. A deployment begins with a defined job, such as resolving a billing issue or handling an employee IT request. OpenAI says the agent receives only the knowledge and system access needed for that job, while the customer defines approved actions, policies, escalation rules and situations that require a person.
The product includes simulations, graders, guardrails and evaluation tools intended to test whether an agent reaches the correct outcome, follows policy, uses tools properly and escalates when required. After launch, production sessions and escalations can reveal gaps. A Codex-powered process then proposes changes, but customer teams test and approve those changes before a controlled rollout.
The reported results need context
OpenAI says Presence operates its English-language phone-support channel and now resolves 75% of inbound issues without human assistance. It also reports that the Codex improvement loop reduced human handoffs by 15 percentage points in 10 days. Help Net Security separately describes the product and correctly attributes those figures to OpenAI. CX Today adds governance context around system permissions and continuous testing.
The retrieved evidence does not disclose the support workload, benchmark design, starting handoff rate, error rate, customer-satisfaction result or independent audit behind those metrics. They are useful deployment claims, but not yet a general benchmark for enterprise agents. BBVA, SoftBank and IAG are described as exploring or testing Presence, which should not be read as evidence of broad production adoption.
What enterprise teams should evaluate
Presence addresses a real deployment gap: capable models need narrow permissions, repeatable evaluations, escalation paths and controlled change management before they can act across business systems. Buyers should inspect those controls at the workflow level.
A serious evaluation should measure task success and policy violations together, test adversarial and ambiguous requests, verify that tool permissions follow least-privilege rules, and confirm that every proposed behavioral change can be reviewed, compared with the production version and rolled back. Limited availability also means implementation quality may depend materially on OpenAI and its integration partners, not only on the product's software components.
Key Points
- 1OpenAI Presence is a limited-general-availability enterprise product for deploying voice and chat agents with scoped access, policies, guardrails, evaluations and escalation rules.
- 2OpenAI reports a 75% autonomous-resolution rate for its own English-language support line and a 15-percentage-point handoff reduction in 10 days, but the retrieved evidence does not independently benchmark those figures.
- 3Codex can propose post-launch changes from production signals, while customer teams test and approve updates before controlled rollout.
Scoring Rationale
Presence combines agent deployment, governance and controlled improvement in a product aimed at consequential enterprise workflows. Its limited availability and vendor-reported performance metrics constrain the score until broader production evidence and independent evaluation are available.
Sources
Primary source and supporting public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
