OpenAI Presence: What Enterprise Voice Agents Cost You

OpenAI launched Presence on July 22, an enterprise platform for deploying AI voice and chat agents that resolved 75% of inbound calls on OpenAI's own support line without human intervention. The platform, which is not self-serve and requires enterprise access starting at $10 million, bundles guardrails, escalation rules, and a Codex-powered improvement loop that reduced human handoffs by 15 percentage points within 10 days on OpenAI's support line. Early adopters include BBVA, IAG, and SoftBank.

OpenAI launched Presence on July 22 — an enterprise platform for deploying AI voice and chat agents bundled with guardrails, company policies, escalation rules, simulation tooling, and a Codex-powered continuous improvement loop. The headline result: OpenAI’s own English-language phone support line now resolves 75% of inbound calls without a human. The fine print: you cannot sign up and start building. Presence is not a self-serve product https://openai.com/index/introducing-openai-presence/ . What Presence Actually Is Each Presence deployment is scoped to a specific job — handling billing disputes, processing insurance claims, answering IT service requests, or running HR inquiries. The agent is provisioned with least-privilege access: only the company systems and data relevant to its defined role. Deployments are led by OpenAI’s Forward Deployed Engineers and select systems integrators. Eligible enterprises can request access; everyone else cannot. The platform bundles several components that developers building on the Realtime API have to assemble themselves: company policies and standard operating procedures baked into agent behavior, guardrails that intervene when a conversation drifts out of scope, escalation rules that route to a human when needed, and pre-deployment simulations that stress-test the agent against edge cases and high-risk scenarios before any real user touches it. The Codex Improvement Loop The most technically interesting piece is what happens after launch. Presence does not require engineers to rewrite prompts when agent behavior drifts. Instead, Codex — OpenAI’s coding agent — reviews production sessions and escalations, then proposes behavioral improvements. Staff approve the changes before they go live. On OpenAI’s own support line, this loop reduced human handoffs by 15 percentage points within 10 days of deployment. That is agentic DevOps: a coding agent iterating on the production agent based on real conversations, without a developer touching the prompt. Early adopters include BBVA, IAG, and SoftBank https://venturebeat.com/orchestration/openai-unveils-presence-a-new-platform-that-lets-enterprises-launch-and-manage-realtime-voice-agents-and-chatbots — banking, insurance, and telecom, which are exactly the sectors where a single agent saying the wrong thing on a live call is a legal problem, not just a UX issue. The Developer’s Real Decision If you are building enterprise voice agents today, Presence clarifies the fork in the road. The Realtime API https://www.open.cx/blog/openai-realtime-api-voice-agent-guide-2026 path costs roughly $0.06 per audio minute and gives you full control. It also gives you full responsibility: carrier integration, tool layer, observability, compliance posture, escalation rules, and the ongoing work of keeping agent behavior aligned with company policy. The audio path is also not HIPAA-eligible under standard Business Associate Agreements as of this writing — a hard stop for healthcare and some financial use cases. The Presence path means OpenAI’s team handles deployment and ongoing tuning. You get compliance-grade infrastructure, a tested simulation framework, and Codex-driven continuous improvement. You give up self-serve access, pricing transparency, and the ability to swap models or providers without renegotiating a contract. The enterprise entry point via the OpenAI Deployment Company starts at $10 million https://thenewstack.io/openai-presence-enterprise-agents/ . The decision is not about model quality — GPT-5.6 is available to everyone. It is about who owns accountability when the agent mishandles a caller’s insurance claim at 2 AM. What This Signals Presence, combined with the Deployment Company formed via the Tomoro acquisition in May 2026 , marks OpenAI’s clearest pivot from API provider to managed AI services firm. The competition is no longer Anthropic or Google — it is Palantir and Accenture. Sub-one-second voice latency is table stakes in 2026. Every serious platform hits it. The actual enterprise differentiator is who guarantees the agent’s behavior, provides the audit trail, and picks up the phone when a compliance team has questions. Presence is OpenAI’s answer to that question. Whether it is the right answer for your organization depends almost entirely on whether your use case requires a $10M contract with OpenAI or whether a well-architected Realtime API deployment with your own guardrails stack gets you to the same place for less. If you are at an enterprise evaluating AI-powered contact center technology, Presence is worth a serious look — request access, demand pricing clarity, and compare it against Rasa, Cognigy, and PolyAI before signing. If you are at a startup or mid-market company, the Realtime API is still your path. Build the guardrails yourself, use Vapi or Retell for orchestration, and keep your options open.