OpenAI introduced Presence on July 22, 2026, a limited-GA enterprise product for deploying voice and chat agents. The notable QA angle is that testing, policy enforcement and controlled improvement are part of the product rather than an afterthought.

What OpenAI Presence includes

Presence combines model reasoning with narrowly scoped system access, company policies, guardrails, approved actions and human-escalation rules. Before launch, teams can run simulations and graders against routine requests, edge cases and higher-risk scenarios.

  • Checks for the right outcome, correct tool use, policy compliance and appropriate escalation.
  • Production sessions, escalations and quality signals that expose gaps after release.
  • A Codex-powered improvement process that proposes changes for teams to test, approve and roll out.

Why this matters for QA engineers

Presence reflects a shift from testing a chatbot’s answers to validating an operational system. QA teams need scenario coverage for tool calls, permissions, policy decisions, handoffs and channel-specific behavior across voice and chat. Regression suites should also compare proposed agent changes with the production version before rollout.

For teams evaluating similar agent platforms, useful release gates include task success, unsupported-action rate, incorrect-tool-call rate, escalation precision, policy violations and performance by intent. Human review remains essential for high-risk workflows and unexpected production behavior.

Availability and limits

Presence is available now to eligible enterprise customers through a limited general availability program led by OpenAI Forward Deployed Engineers and selected systems integrators. It is not a self-serve product. OpenAI says its own English-language phone-support deployment resolves 75% of inbound issues without human assistance, but that company-reported result should not be treated as a benchmark for every workflow.

Sources