OpenAI's $150 million enterprise push puts forward deployed engineers inside client operations

OpenAI has committed $150 million to a new enterprise partner program and formed a standalone services company, the OpenAI Deployment Co., staffed with engineers who work on-site inside client organizations. The moves, reported by The Wall Street Journal on July 22, mark the company's most direct attempt yet to close a gap that has frustrated enterprise technology buyers for two years: owning an AI model license is not the same as having AI that works.

Two programs, one problem

The $150 million partner program and DeployCo run in parallel but serve different ends. The partner program extends OpenAI's reach through a network of third-party implementers. DeployCo, by contrast, is a direct-delivery firm designed to put OpenAI's own technical talent inside the customer's environment.

Both rely on forward deployed engineers, a staffing model that has become common among AI-native vendors precisely because enterprise deployments tend to stall when left to a customer's internal team alone. The workflow involves FDEs co-developing applications alongside the client's staff, then handing off a working system rather than documentation.

Arnaud Fournier, who serves as chief technology officer of DeployCo and is an OpenAI veteran, told the Journal that there has never been a greater gap between what AI models can do and what enterprises are actually using them for. That framing shapes the entire service model: DeployCo is not selling consulting advice but production engineering capacity.

Built through acquisition

DeployCo launched in May 2026 with the acquisition of Tomoro, an AI consulting and engineering firm that brought roughly 150 engineers into the organization. A subsequent acquisition of applied AI firm Northslope pushed the FDE headcount into the hundreds, according to the Journal. A small group of engineers from OpenAI's own internal FDE organization is also on temporary loan to help DeployCo scale.

The firm is still selecting its first customers and is prioritizing organizations with existing OpenAI relationships. BBVA, the Madrid-based bank, offers an early illustration of the model in practice. Fran Martínez López, head of risk transformation at BBVA, told the Journal that an OpenAI engineer helped his team build a new AI application for credit risk analysis, a project the bank had been working on for months before the FDE engagement accelerated it. The work is part of a broader effort to deploy AI and agents across BBVA's operations.

The service approach matters in the context of OpenAI's broader business pressure. Enterprise revenue has become a critical segment as the company works toward justifying heavy infrastructure investment and a path to public markets, according to the Journal. At the same time, business leaders are pushing back on AI costs and questioning whether deployments are delivering measurable value, which is exactly the concern DeployCo is structured to answer.

What the AI labs' own offices reveal

The internal practices at OpenAI, Google, and Anthropic offer a concrete signal of where enterprise workflows are heading. Reporting by the Journal in June found that while most organizations are still using AI for narrow tasks like meeting transcription or memo drafting, these companies are assigning more complex, multistep work to agents that operate across applications.

Even nontechnical departments at these firms are running agents for tasks that previously required human execution at every step. The structural shift is that employees become reviewers of AI output rather than primary producers, effectively fact-checking and approving work the agent has already done. The Journal noted that agent deployments have not been without friction: incidents including mass email deletions and disappearing code have occurred when agent behavior was insufficiently scoped.

That tension between capability and reliability is exactly what the FDE model is meant to manage. An on-site engineer who understands both the AI system and the client's operational environment can define guardrails, set scope boundaries, and monitor outputs in ways that a remotely purchased API subscription cannot.

What this means for your team

• Evaluate vendor service depth, not just model access: when assessing AI platforms for renewal or expansion, ask whether forward deployed engineering capacity is available and how engagements are scoped. A partner who can staff on-site is a different procurement decision than a SaaS license.
• Audit your current AI deployment gaps: if internal teams have been sitting on model access without shipping production applications, DeployCo's prioritization of existing OpenAI customers means you may be eligible for early engagement. Identify two or three high-value workflows to bring to that conversation.
• Plan for a reviewer-not-producer workforce model: the agent adoption pattern at OpenAI, Google, and Anthropic suggests that job scopes shift toward verification and exception handling. IT and operations leaders should begin mapping which roles will absorb that change and what tooling or training they will need.
• Define agent guardrails before deployment, not after: the Journal's reporting on agent mishaps at AI-native companies underscores that scope boundaries need to be engineering decisions, not policy memos. Include guardrail architecture in any FDE engagement brief.