OpenAI has introduced OpenAI Presence, a new enterprise product designed to help organizations deploy AI agents for customer service and internal operations across voice and chat. The company is positioning the launch as a response to a practical shift in the market: for large companies, the main challenge is no longer whether AI agents can produce useful answers, but whether they can operate reliably inside real workflows with permissions, policies, escalation paths, and continuous updates.

According to OpenAI’s official announcement, Presence is not a self-serve software release. It is being offered through a limited general availability program for eligible enterprise customers, with deployments led by OpenAI Forward Deployed Engineers and select systems integrators. That positioning matters. Rather than selling a general-purpose toolkit alone, OpenAI is packaging models, workflow controls, evaluations, guardrails, and deployment support into a managed product for higher-stakes use cases.

For AI builders and enterprise buyers, the news is less about a brand-new model and more about OpenAI moving further up the stack. Presence suggests the company sees the next competitive layer in enterprise AI as operational reliability: connecting models to company systems, constraining behavior, testing edge cases, and improving agents after launch without handing over full autonomy.

What Presence is supposed to do

OpenAI describes OpenAI Presence as a platform for “trusted AI agents” that can answer questions, resolve issues, use company systems, take approved actions, and hand work off to people when needed. The company says each deployment begins with a narrowly defined job, such as billing support, insurance claims help, or employee IT service requests.

In OpenAI’s description, the agent only receives the knowledge and system access needed for that specific task. The customer sets the operating rules: which actions are allowed, when approval is required, and when a human must take over. Presence then combines those rules with model reasoning, guardrails, evaluation systems, and escalation logic.

That framing puts Presence closer to an enterprise operations layer than a standalone chatbot. OpenAI says the product includes policies and standard operating procedures, approved actions, simulations, evaluation tools, guardrails, and a Codex-driven improvement loop. The idea is that enterprises can keep some controls consistent across multiple deployments while changing workflow-specific details by department, channel, or use case.

The company says Presence currently supports real-time experiences in voice and chat, including customer support, outbound sales, and high-risk internal workflows. OpenAI’s examples include verifying users, retrieving account information, applying company policy, and completing approved actions.

Why OpenAI is packaging this as a managed enterprise product

The launch reflects a broader enterprise AI reality: deploying agents in production has become less about demo quality and more about controlled execution. Many organizations already know modern models can draft responses or navigate simple support flows. The harder problem is keeping those systems accurate and compliant as products, policies, and user behavior change.

OpenAI’s message is that Presence addresses that problem by combining model capability with operational structure. The company says production sessions and escalations reveal gaps after launch, and that Codex can propose updates that teams then test and approve. That is a notable detail because it positions Codex not just as a coding assistant, but as part of a maintenance loop for deployed agents.

OpenAI also says the product was developed in collaboration with its research organization and shaped by years of customer deployments. It argues that insights from each implementation feed back into product and research improvements. That may help explain why Presence is launching first as a guided service rather than a broad self-serve offering: OpenAI appears to be treating deployment knowledge itself as part of the product.

The involvement of OpenAI Forward Deployed Engineers reinforces that point. OpenAI says its own teams work with customers to identify high-value workflows, connect internal systems, establish permissions and policies, test the agent, and bring it into production. For buyers, that may reduce implementation risk. For developers hoping for a plug-and-play platform, it also signals that Presence is currently more consultative than software-only.

Early deployments and the performance claims

OpenAI’s strongest proof point is its own internal use of Presence for the company’s English-language phone support line at 1-888-GPT‑0090. According to OpenAI, that deployment handles open-ended requests, verifies callers, uses account context, and takes approved actions. The company says that within weeks it met or exceeded the benchmarks it uses to evaluate frontline human support quality.

OpenAI also reports that the system now resolves 75% of inbound issues without human assistance, and that a Codex-powered improvement loop reduced human handoffs by 15 percentage points in 10 days. Those are meaningful numbers if they hold up in broader customer settings, especially because support automation often struggles once identity checks, policy rules, and system actions are required.

But those metrics are vendor-reported, and the company has not provided underlying methodology, sample size, benchmark definitions, or independent validation in the source material provided here. That does not make the claims false; it means outside observers should treat them as OpenAI’s own operating evidence rather than audited market data.

OpenAI also names three enterprise organizations as early participants. BBVA is exploring AI-powered voice support for everyday banking needs in Mexico. SoftBank is testing natural Japanese-language customer conversations. IAG is exploring support during high-demand events such as severe weather. Those references help show market interest across finance, telecom, and travel-related operations, but OpenAI does not disclose deployment scale, production status, commercial terms, or outcome metrics for those organizations.

Evidence, boundaries, and what is still unclear

Because the source cluster here is entirely OpenAI-controlled, most factual detail comes directly from the company’s own announcement. That means the product scope, customer examples, and benchmark numbers all rely on vendor reporting.

What appears confirmed from the announcement is that OpenAI Presence exists as a limited-availability enterprise offering; it supports voice and chat workflows; it is deployed with the help of OpenAI Forward Deployed Engineers and selected partners; and it is not yet a self-serve product. OpenAI also clearly states that standard OpenAI API access for voice use cases will continue separately.

Several important points remain unclear. OpenAI has not disclosed pricing, eligibility criteria for the limited general availability program, implementation timelines, model configurations, region availability, or technical integration requirements. The company also does not say whether Presence is built as a layer over standard API components or whether some capabilities are exclusive to managed deployments.

There is also a distinction between exploration and production that buyers should note. OpenAI says BBVA and IAG are “exploring” certain use cases and that SoftBank is “testing” conversations. Those words suggest early-stage work, not broad operational rollout.

What this means for enterprise AI teams

For enterprise AI teams, OpenAI Presence is a signal that the competition around enterprise AI is shifting from raw model access to end-to-end deployment systems. Many organizations do not just need a strong model; they need repeatable controls for identity verification, approved tool use, policy compliance, quality testing, and escalation.

That is especially true in customer support and internal service desks, where errors are measurable in cost, compliance exposure, and customer frustration. If Presence performs as OpenAI claims, it could appeal to enterprises that want faster deployment than a custom build but more control than a generic conversational bot.

The tradeoff is likely flexibility versus dependence on OpenAI’s service layer. Buyers considering OpenAI Presence will need to weigh the benefit of a battle-tested managed product against questions about portability, pricing, and how much customization remains under their direct control. Since Presence is not self-serve, the product may be best suited initially to large organizations with defined workflows and the budget for guided deployment.

For builders, the use of Codex inside an improvement loop is one of the more interesting product signals. It suggests OpenAI sees agent maintenance as a continuous engineering task: inspect real sessions, identify failure patterns, propose changes, test them against the live version, and then approve rollout. That is a more operational view of AI agents than the simple prompt-and-deploy model that dominated early experimentation.

What to watch next

The next important signal will be whether OpenAI expands OpenAI Presence beyond limited general availability and turns it into a broader platform with clearer packaging. Pricing, deployment speed, supported integrations, and customer case studies will determine whether Presence remains a high-touch service for select accounts or becomes a scalable enterprise standard.

A second signal is independent evidence. If BBVA, SoftBank, or IAG eventually publish measurable outcomes, that would give buyers a better basis for comparison than OpenAI’s own benchmarks alone. Enterprises will also want to see how Presence performs in non-English and regulated settings, where escalation design, auditability, and localization matter as much as model fluency.

Third, watch the relationship between Presence and the OpenAI API. If OpenAI keeps advanced operational controls mainly inside managed deployments, it could create a two-track strategy: self-serve model access for developers, and higher-margin managed systems for enterprise AI programs.

Creati.ai perspective

OpenAI Presence looks like a deliberate move from model provider toward managed enterprise operator. The key message is not that OpenAI has another agent demo; it is that the company wants to own the layer where models become governed workflows tied to real systems and business rules.

That matters because enterprise buying decisions are increasingly being made on reliability, control, and operational support rather than model novelty alone. If OpenAI can prove that Presence consistently reduces handoffs, enforces policy, and speeds deployment in environments like customer support, it will strengthen its position against both horizontal AI platforms and service-heavy integrators. For now, though, the market has mostly OpenAI’s own evidence. The product is promising, but the next stage of the story will depend on independent customer outcomes and whether OpenAI can turn a high-touch service into a repeatable platform.