OpenAI is turning AI agent deployment into a consulting product, and it is doing it with a very specific constraint: Presence is “not yet available as a self-serve product.” The company said Presence supports real-time experiences across voice and chat, including customer support, outbound sales, and high-risk internal workflows, plus an agent editor, agent playground, and simulations for tasks like customer refunds, order status queries, account deletion, and other administrative interactions. In other words, this is OpenAI selling the plumbing, but it is selling it the way enterprise IT buys risky systems: with guidance, governance, and hands-on deployment.

Presence is positioned for scenarios where the cost of getting things wrong is high, so OpenAI built in policy and governance controls that shape how agents behave. The company’s example is a billing issue resolution workflow: understanding the request, verifying the customer, looking up account information, applying company policy, and taking an approved action. OpenAI also says the control interface lets enterprises simulate how an agent handles those admin tasks before letting it loose. If the request needs human support, the system can bring in a person. And for anyone wondering whether this is mostly a technical demo, OpenAI claims it has been dogfooding Presence, which now runs its AI English phone support channel.

So why does it matter that Presence is “available through its consulting arm” instead of being released like another self-service API? Because it suggests OpenAI is matching the messy reality of enterprise adoption. AI models may be getting more commoditized, but deployment is where budgets live and where liability shows up. Enterprises are not only buying model capability, they are buying operational assurances: who can do what, under which conditions, with what audit trail, and what happens when a model hits an edge case. Presence’s emphasis on governance controls, simulation tooling, and approved actions reads like a direct response to that procurement reality.

This also shows up in the way OpenAI is packaging the workforce behind it. Deployments, the company said, are led by OpenAI Forward Deployed Engineers and select global systems integrators. Forward Deployed Engineers debuted last year and was bolstered by OpenAI’s acquisition of consultancy Tomoro in May. That matters because it frames Presence as something OpenAI intends to install, not just sell. The company’s approach is basically: if customers want agents that can touch customer data, trigger business processes, and potentially resolve refunds, then OpenAI wants to own the rollout experience end-to-end, at least in the early stage.

OpenAI is also being careful about how widely it rolls the product out. During this “limited GA phase,” deployments are scoped individually based on each customer’s use case and implementation needs. Broader pricing details are “to come as availability expands,” and Presence is available to eligible enterprise customers. In plain English, OpenAI is avoiding the universal “here’s the price, start plugging in tomorrow” moment. If you are trying to convert enterprise skepticism into signed contracts, scoping is a way to control risk, manage implementation variance, and learn how different customers want policy and governance to work.

The market context matters. Gartner predicted last month that by 2027, half of organizations planning to shift customer service to AI will abandon those plans. Gartner’s Kathy Ross, senior director analyst for the customer service and support practice, said AI is “not a panacea,” adding that “The human touch remains irreplaceable in many interactions.” Presence reads like OpenAI trying to beat that skepticism with something more enterprise-friendly than a generic chatbot: agent behaviors constrained by policy, simulations for administrative workflows, and a built-in human handoff when support is required.

And then there is SoftBank, which looks like a customer OpenAI really wants in the proof column. SoftBank has committed $60 billion to the AI biz, and through its collaboration with OpenAI, it is exploring how Presence can enable “trusted customer agents” that communicate naturally and represent SoftBank consistently across customer interactions. SoftBank’s statement to The Register included remarks from Tadahisa Murakami, VP and head of SoftBank’s data and digital transformation division, who said frontline teams rated the agent’s Japanese-language conversations highly for their natural and accurate quality.

Presence is not just about feature parity with other agent approaches. It is also about unit economics and enterprise spending power. OpenAI’s AI strategy still has to cover the costs of its massive datacenter buildout commitments while it works toward eventual profitability, and an “actual money” product helps. With Presence, OpenAI can charge for deployment work, implementation, and integration. That is a different revenue profile than pushing more self-serve usage, and it fits the “margin in the plumbing” logic: models can be commoditized, but systems that safely execute business tasks are harder to swap.

For executives and boards evaluating AI investments right now, the strategic stake is simple: model access is no longer the differentiator. Deployment ownership, governance controls, and credible enterprise rollout paths are becoming the differentiator. If Presence proves out, it could set a pattern for how frontier model providers monetize the enterprise shift: not by launching another checkbox API, but by offering controlled agent installations with a workforce behind them. If it fails, it will likely reinforce Gartner’s warning that many customers will stall out when AI touches customer interactions and high-risk workflows without enough operational reassurance. Either way, OpenAI is betting that the enterprise market will pay for certainty, not just intelligence.