On July 22, OpenAI introduced Presence - a new enterprise product that is designed to help deploy and manage AI agents across customer-facing and internal business operations workflows.
This launch marks the company’s strategic move to go beyond model wars and expand into the broader enterprise software market. It seems reasonable - after making AI popular with users who mostly prefer free or low-cost tools, OpenAI is now focusing on enterprise customers who can spend more for services they need.
The Enterprise Problem Presence Is Trying to Solve
The solution is aimed at corporate clients and addresses the main barrier that hinders the adoption of AI agents in enterprise environments. It’s controllability.
An agent who can only answer background questions is one type of risk. AI agents operating in environments where mistakes are no longer harmless chatbot errors but operational or reputational risks are a completely different challenge. In a bank or insurance group, an agent’s mistake can cause financial losses, damage trust, and create regulatory issues. In internal IT or HR processes, an agent must not only respond politely, but also act in accordance with roles, permissions, and corporate policies.
So, an agent capable of performing real-world actions like changing account data, approving an action, creating a request, or working with financial information requires a different level of control. To be effective, they need more access rights and corporate context. But at the same time, they also need to strictly adhere to specific corporate policies and evaluation standards. On top of that, different companies, like contact centers, banks, and insurance companies, all have different requirements, which further complicates their implementation.
And that's exactly what Presence is trying to sell: not a smarter model, but the control layer enterprises need to trust AI agents with real business tasks.
How Presence Works in Practice
Presence is expected to help companies launch an agent with clearly defined responsibilities. According to OpenAI, Presence agents can conduct real-time voice and chat conversations in customer support, sales, IT requests, and other repetitive business processes. It could be checking a customer account, responding to a typical insurance request, processing an internal request to the IT department, or initial communication with a potential buyer. The Presence agents do not get access to all the company's systems, but only to the data and actions needed for a specific scenario.
They answer questions, verify the caller's identity, retrieve account information, enforce company policies, and perform approved actions, like refunds or invoice adjustments. When a request exceeds the mandate, the platform routes the request to a human operator. The company itself can determine which actions an agent can perform independently, which require approval, and when control must be delegated to a human, with safeguards intervening if interactions cross established boundaries.
So, generally, it’s more of a deployment platform than a separate model: at the heart of the product is not a new benchmark, but a set of rules and processes that should make agents suitable for real work where you need access rights, policy compliance, logging, a clear escalation procedure, quality control, and the ability to quickly correct an agent if they start making mistakes.
Beyond an API: OpenAI's Enterprise Deployment Strategy
What is important is that Presence is not another self-service API, at least for now. Instead, it is a boots-on-the-ground project with customized scope, customized pricing, and human implementation.
To make it possible, OpenAI has been consciously building its consulting arm - the OpenAI Deployment Company. In May 2026, it acquired consulting firm Tomoro, bringing approximately 150 deployment and implementation engineers into the company. So, all Presence deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. OpenAI and its implementation partners help companies decide which tasks to automate, integrate AI with their existing systems, configure the AI's permissions, and thoroughly test it before deploying it for real customers.
Continuous Evaluation and Improvement
OpenAI also relies on Codex to improve agents. Once a production session is running, escalations and quality signals can show where an agent has failed or where policies have become outdated. Changes don’t have to automatically roll out to users: the team can review
suggestions, test them, and only then agree to a controlled update.
The key idea is to monitor before, during, and after launch. Before public use, a company can run an agent through simulations, typical queries, edge cases, and high-risk scenarios. The checks should show whether the agent follows policies, uses tools correctly, does not take on prohibited actions, and transfers the conversation to a human in a timely manner.
Availability, Pricing, and Early Customers
OpenAI Presence is available to eligible enterprise customers as a deployed product through a limited general availability program. Pricing for Presence itself has not been disclosed. Right now, the product is only being sold to selected customers with customized pricing, but broader pricing will be announced later.
The launch materials mention BBVA, SoftBank, and IAG as companies that are studying or testing Presence. Spanish financial institution BBVA is exploring AI-powered voice support for everyday banking needs in Mexico. Telecommunications giant SoftBank is testing natural customer conversations in Japanese, and insurance group IAG is evaluating a system to provide timely assistance during peak times, such as foul weather or insurance events.
SoftBank already says the AI performs well in Japanese conversations. This is good news for OpenAI because enterprise customers generate substantially higher revenue than consumer subscriptions, which OpenAI needs to cover its huge infrastructure costs. It’s equally important for companies because OpenAI is no longer selling just a model or chatbot, but a full layer of control: permissions, policies, checks, escalations to humans, and a process of continuous improvement of agents after launch.
OpenAI also claims that it uses a Presence agent on its own English-language telephone support channel at 1-888-GPT-0090. Within weeks of deployment, it met or exceeded internal benchmarks for frontline human support quality and now autonomously resolves approximately 75% of incoming issues, and recent improvements have reduced the need to transfer calls to people. However, these results were reported by OpenAI itself and haven't been independently
verified.
Final Thoughts
OpenAI is effectively moving towards a consulting and integration model, where value is created not just by the model itself, but also by how it is embedded in the client’s work. It’s a logical step as basic AI models become increasingly affordable and competition between OpenAI, Google, Anthropic, Meta and others squeezes margins. The Presence launch suggests that the next phase of enterprise AI competition is no longer just about building smarter models. It's about deploying AI safely, reliably, and at scale in real business environments. For OpenAI, it also opens a new path to monetizing enterprise AI beyond selling access to its models.
