OpenAI has unveiled Presence as an enterprise product for phone and chat support channels, with reporting on 2026-07-22 describing it as a platform for deploying AI agents into production support workflows. The launch targets large enterprises evaluating service automation, but the trust case still depends heavily on OpenAI’s own operating claims and a restricted, services-led rollout.
What changed
The New Stack reported that OpenAI unveiled Presence on 2026-07-22 for enterprise phone and chat channels. Multiple sources describe Presence not as a standalone model but as a deployment platform for defining, deploying, evaluating, and improving AI agents.
OpenAI says the product uses the same support agents it runs on its own English-language phone support line, which No Jitter identified as 1-888-GPT-0090. According to OpenAI, that internal support agent currently resolves 75% of inbound issues on the English-language phone line without human assistance.
Across the reporting, Presence is described as handling specific support tasks that go beyond answering questions. OpenAI says the agent can verify identity or callers, look up accounts or use account context, resolve billing issues, make account changes, process refunds, and take approved actions directly.
Help Net Security and No Jitter, citing OpenAI, described Presence as combining policies, SOPs, guardrails, approved actions, simulations, evaluation tools, monitoring, and a Codex-powered improvement process.
Why B2B teams should care
One of OpenAI’s central enterprise arguments is its own internal use of the platform: it is asking buyers to trust a support system that it says already runs in its own operation. That is a stronger sales narrative than a generic model launch, but it still depends heavily on OpenAI claims rather than independent verification.
For CTOs and CX leaders, the more important detail is that, across the available reporting, Presence is positioned as an operating layer around the agent rather than just another model. The New Stack reported that Presence bundles the agent with policies, testing tools, and monitoring.
The reported control set is concrete. The New Stack said teams can simulate policy changes against past cases before rollout. Help Net Security said teams can test agents against common requests, edge cases, and higher-risk scenarios. The New Stack also reported a live dashboard for response accuracy, call volume, and task handling.
Who is affected
Customer service is the lead use case, but OpenAI’s stated scope is wider. No Jitter reported that OpenAI is positioning Presence for customer support, sales development, procurement, IT, HR, and internal service requests.
Availability is constrained. The New Stack and No Jitter both reported that Presence is available only to eligible enterprise customers through a restricted rollout or limited general availability program, and Help Net Security reported that it is not a self-service product.
The named references are early, not broad-market proof. The New Stack cited BBVA, SoftBank, and IAG as design partners, while No Jitter named BBVA Mexico, SoftBank Corp., and Retail Insurance Australia, IAG, as early customers cited by OpenAI.
The New Stack reported that deployments are handled by OpenAI forward deployed engineers and systems integrators, while No Jitter said OpenAI and partners help connect systems, establish permissions, test agents, and bring them into production.
What teams should check now
Enterprise buyers evaluating OpenAI support agents for enterprises should pressure-test the operating model before treating the launch as a software purchase.
- Confirm whether the agent can be restricted to approved actions only, especially for refunds, account changes, and other transactional tasks.
- Ask how guardrails intervene when interactions move outside company boundaries, using OpenAI’s described guardrail model as the baseline.
- Require historical-case simulations before rollout, since The New Stack reported that policy changes can be tested against past cases.
- Verify whether pre-production testing covers common requests, edge cases, and higher-risk scenarios.
- Confirm which operational metrics are available, including response accuracy, task handling, and escalation visibility.
- Ask for the escalation design and human handoff assumptions, because Help Net Security reported that production sessions and escalations are used to improve behavior after launch.
- Ask about language coverage and channel scope, because OpenAI’s strongest internal proof point is its English-language phone support line.
- Clarify implementation requirements and staffing, because Presence is supported by forward deployed engineers and partners rather than sold as a self-service tool.
- Clarify procurement assumptions because pricing has not been disclosed; for general evaluation criteria, see our SaaS Pricing Models: How Buyers Should Evaluate Cost.
What remains unclear
- Not yet confirmed: pricing for Presence, beyond OpenAI’s statement reported by No Jitter that broader pricing details will come as availability expands.
- Not yet confirmed: independent verification of key performance and customer claims, including OpenAI’s 75% resolution figure and cited customer quotes.
- Not yet confirmed: the exact eligibility boundaries for the restricted rollout or limited general availability program.
- Not yet confirmed: whether Presence is production-ready across all customer service use cases, industries, languages, or regulatory environments.
- Not yet confirmed: how much customization customers can perform without OpenAI’s forward deployed engineers, the OpenAI Deployment Company, or systems integrator partners being involved.
What to watch next
The next meaningful signals are broader availability, pricing disclosure, named enterprise deployments beyond the initial partner set of BBVA, SoftBank, and IAG, and evidence that OpenAI expands Presence from support into IT, HR, procurement, and sales development.
No Jitter also reported an OpenAI claim that Codex reduced human handoffs by 15 percentage points in 10 days.
