OpenAI Unveils Presence for Enterprise Voice and Chat Agents

OpenAI has taken another step toward owning the full enterprise AI agent stack. On July 22, 2026, the company announced Presence — a new platform that bundles agent deployment, policy enforcement, evaluation, guardrails, and post-launch monitoring into a single managed offering. It is available immediately through a limited general availability program.

The pitch is straightforward: if your company wants AI agents answering phones, handling chats, and interacting with internal systems, Presence promises to handle the messy plumbing that makes those agents reliable in production. But there is a catch — and it is one that every team building serious AI workflows should think carefully about.

What Presence Actually Packages

Presence is not a model. It is an infrastructure layer that sits around OpenAI's models and assembles everything an enterprise needs to run agents that behave predictably in the real world.

According to OpenAI, the platform brings together company knowledge, standard operating procedures, approved actions, simulations, evaluation tools, guardrails, and escalation rules. Each deployment starts with a defined job — resolving a billing issue, supporting an insurance claim, handling an employee IT request — and the agent receives only the information and system access required for that specific task.

The customer determines what the agent may do independently, which actions require approval, and when a human must take over. Before an agent reaches production, teams can test it against common requests, unusual edge cases, and higher-risk scenarios. Graders evaluate whether the agent reached the intended outcome, followed policy, used tools correctly, and escalated when required.

After launch, Presence monitors production sessions, escalations, and quality signals. A Codex-powered improvement loop investigates those signals and proposes updates. Teams then test a proposed change against the version already in production before approving a controlled rollout — a mechanism designed to prevent an agent from silently drifting as policies or user behavior shift.

This is a genuine problem. The gap between a demo that impresses executives and an agent that handles thousands of live interactions without embarrassing the company is enormous. OpenAI is betting that enterprises will pay to close that gap rather than building the infrastructure themselves.

How Deployments Work — and Who They're For

Presence is not a self-service product. OpenAI Forward Deployed Engineers (FDEs) and select global systems integrators lead deployments. There is no sign-up form, no free tier, no API key you plug in on your own.

This immediately narrows the audience. FDE-led deployments are resource-intensive on OpenAI's side, which means Presence is aimed at large enterprises with the budget and patience for a consultative sales process. Pricing has not been disclosed. Geographic limits, contractual terms, and the expected cost of engineering and integration work have not been disclosed either. OpenAI did not respond to repeated requests for clarification on these points.

The product is available today for real-time voice and chat experiences. OpenAI's materials describe broader ambitions spanning voice, chat, email, and other channels, but email support has not been confirmed at launch.

On the model question, an OpenAI spokesperson clarified that Presence uses OpenAI models for the core agent, while allowing customers to connect third-party models and services through APIs for guardrails, tools, and other parts of their workflow. In other words, the brain is OpenAI. You can bolt on other pieces, but the decision-making core is not interchangeable.

Early Results — and the Fine Print

OpenAI is eating its own cooking. Presence powers the company's English-language phone-support channel at 1-888-GPT-0090, where the system handles open-ended requests, verifies callers, uses account context, and performs approved actions. According to OpenAI, it now resolves 75% of inbound issues without human assistance. A Codex-powered improvement loop reduced human handoffs by 15 percentage points over a 10-day period.

Those are company-reported figures that have not been independently verified, and they come from OpenAI's own support line — a controlled environment where the company has perfect knowledge of its own systems and policies. The real test is what happens when Presence meets the complexity of a bank's legacy infrastructure or an insurer's crisis workflows.

That said, early evaluators are significant. BBVA is exploring voice support for routine banking needs in Mexico. SoftBank is testing natural Japanese-language customer conversations. Australian insurer IAG is exploring support during high-demand periods such as severe weather and natural disasters. These are serious organizations with serious compliance and reliability requirements. Whether Presence delivers for them will be the real proof point.

The Managed-Platform Trade-off

Presence solves a real problem, but it also illustrates a trade-off that is becoming central to enterprise AI strategy.

When you build on a managed platform, you get convenience and a fast path to production. You also get vendor lock-in, opaque pricing, and limited control over the most critical decisions. OpenAI chooses which model runs your core agent. OpenAI's engineers lead your deployment. OpenAI's evaluation framework defines what "working correctly" means. You operate within the guardrails they provide, and your ability to customize or audit the underlying logic is constrained by what the platform exposes.

For some organizations, that trade-off is acceptable — even preferable. A company without AI engineering talent and with a budget for consultative deployments may be better off handing the problem to OpenAI entirely.

But for teams that want to own their agent infrastructure — who need to control which models run where, how guardrails are implemented, what data leaves their network, and how costs scale — a self-hosted approach offers something Presence structurally cannot: full transparency and full control.

This is not a theoretical concern. When your core agent is locked to one provider's models, switching costs compound over time. When your guardrails and evaluation logic live inside someone else's platform, auditing them for regulatory compliance becomes a question of what the vendor is willing to show you. When pricing is undisclosed and deployments require dedicated engineers, cost predictability depends entirely on the vendor's goodwill.

What This Signals for the Agent Market

OpenAI's move with Presence confirms something the market has been converging on: the bottleneck for enterprise AI is no longer model capability. It is operational reliability — getting agents to behave correctly, consistently, and safely at scale over time.

That bottleneck is driving a wave of managed solutions, from Presence to similar offerings from other providers. Each one promises to abstract away the complexity. Each one also centralizes control in ways that may not serve every team's interests.

The teams that will benefit most from this moment are the ones that think clearly about what they want to own and what they're willing to outsource. Managed platforms like Presence have a role to play, especially for organizations that need a fast, low-effort path to production and have the budget for it.

But for growing teams — startups, agencies, internal ops groups, mid-market companies — that want AI agents as a permanent part of their infrastructure rather than a rented service, the economics and control model of self-hosted solutions are increasingly hard to ignore. You own the runtime, you choose the models, your data stays in your network, and the cost is predictable. Tools like OfficeForge exist precisely for this use case — a team of AI employees on your own server, with a one-time cost and no vendor hand on the steering wheel.

Presence will find its market. The question for every team evaluating it is whether that market includes them — or whether they'd rather build on ground they own.