AINEWS
OpenAI's 'Presence' service sells staffed deployments of AI agents through its consulting arm, signaling that the most resourced lab no longer thinks model margins are the growth story.
OpenAI opened a new line of business on Tuesday. The line is called Presence, and what it sells is not intelligence. It sells engineers.
The service lets enterprises deploy AI agents for voice and chat workflows, including customer support, outbound sales, and what OpenAI calls high-risk internal tasks. Access comes only through the company's consulting arm, the OpenAI Deployment Company, and the Forward Deployed Engineers (FDEs) who staff it, a delivery model that is not available as a self-serve product (The Register).
The product sits on a bench the company has spent the last year building. In May 2026, OpenAI bought UK AI consultancy Tomoro, a long-time enterprise integrator, specifically to staff the deployments it is now selling through Presence (prior The Register coverage). The press framed the launch as a product announcement. The structural read is a market signal: even the best-resourced frontier lab has decided token margins are not where the next leg of growth comes from. The profit center has moved up the stack into deployment labor, governance tooling, and the kind of staffed engagements that look more like Accenture than AWS.
The first named customer sits inside the same capital structure as the vendor. SoftBank is a reference customer using Presence for Japanese-language customer-service agents, according to The Register. SoftBank VP and head of data and digital transformation Tadahisa Murakami told The Register that frontline teams rate the natural-language quality as highly accurate. The other live deployment is OpenAI itself: Presence runs the company's own English-language phone-support channel, handling refund eligibility and account changes with human handoff when the agent cannot close the case (The Register).
Two things are worth noticing. The customers are SoftBank and OpenAI. SoftBank is an investor and is using Presence for Japanese-language customer-service agents. OpenAI is running the product for its own English-language phone-support channel. No independent third party has published a Presence deployment at scale, and pricing is individually scoped per engagement during the limited general-availability window, with broader pricing to follow as availability expands (The Register). Second, a relevant data point predates the announcement. In June 2025, Gartner predicted that by 2027, 50% of organizations planning to shift customer service to AI will abandon those plans, with senior director analyst Kathy Ross saying the human touch remains irreplaceable in many interactions (Gartner). Presence is OpenAI's structural answer: if human handoff is unavoidable, the human part of the engagement is where the vendor will make its margin.
The mechanism is not new in the IT services market. The frontier-lab wrapper is. McKinsey, Accenture, and the Indian services majors sell exactly this model: senior engineers, governance tooling, and a multi-year change-management engagement. OpenAI adds one layer the services majors cannot: the model underneath the engagement. That lets the company price the labor against a benchmark the customer cannot replicate without also buying the underlying API access. For enterprise buyers, the decision is no longer "which model." It is "which lab will staff the engagement, and at what hourly rate." For Anthropic, Google DeepMind, and xAI, the question is whether to follow the same up-the-stack move or hold the line on raw API pricing and let integrators absorb the labor. The falsifier is easy to name. If a rival lab undercuts on raw API price and wins enterprise deals without staffing, then Presence is an OpenAI-specific commercial bet, not a category shift, and the thesis downgrades to a single-vendor story.
Presence includes an agent editor, an agent playground, and a simulation tool for testing policies and governance before any agent talks to a real customer. Those components look like every other enterprise agent platform shipped in the last six months, with one difference: they are bundled into a delivery that arrives with an OpenAI engineer on the call. The product news is the staffing model. The pricing news is the staffing model. The category shift, if there is one, is the staffing model.
The first published rate card, when OpenAI exits limited GA and posts per-engagement or per-FTE pricing, is the test of whether the deployment-services thesis is a real category shift or a one-vendor rebrand. What the press has called "boots-on-the-ground" pricing so far is The Register's characterization, not an OpenAI-stated number.