From Selling Models to Living Inside the Business
For most of the last three years, the enterprise AI relationship was simple. A vendor sold you a model or a copilot, you wired it into your stack, and the hard work of implementation was your problem. OpenAI has just torn up that arrangement. The OpenAI Deployment Company, launched with more than 4 billion dollars of investment at a 10 billion dollar valuation, is built to put OpenAI's own people inside your organization to build the AI for you.
The vehicle for this is the forward deployed engineer, or FDE, a role borrowed from Palantir's playbook and now AI's hottest job. The company acquired consulting firm Tomoro to seed roughly 150 of them from day one, alongside founding partners Bain and Company, Capgemini and McKinsey and Company. As OpenAI Chief Revenue Officer Denise Dresser framed it, "AI is becoming capable of doing increasingly meaningful work inside organizations. The challenge now is helping companies integrate these systems into the infrastructure and workflows that power their businesses." That is a consulting business, and OpenAI is no longer pretending otherwise.
What an Embedded Engineer Actually Sees
The pitch is seductive because the problem is real. Enterprises are drowning in pilots that never reach production, and the gap between a working demo and a governed, integrated system is exactly where most AI value evaporates. An FDE who parachutes in and lives inside the complexity, the legacy infrastructure, the compliance constraints, the convoluted permissions, can close that gap faster than a team shipping software and walking away.
But that intimacy cuts both ways. To build durable systems, an embedded engineer has to see the process debt, the manual workarounds, the brittle APIs, the compliance exceptions, the data quality problems, and the spreadsheet that refuses to die. In other words, they see your operating system in its true, unflattering state. That is enormous value to extract and an enormous exposure to grant, and CIOs are right to feel the tension between the two.
The Trust Question Ownership Does Not Answer
OpenAI says the new company is majority-owned and controlled, a line clearly meant to reassure. Analysts are not fully reassured. Sanchit Vir Gogia, chief analyst at Greyhound Research, put it sharply: "Majority ownership and control give buyers a headline answer, not a complete trust answer." His prescription is the one CIOs should adopt verbatim. "CIOs should not only ask who owns the shares. They should ask who governs the work."
That distinction is the whole ballgame. Ownership is a cap table. Governance is whether you control who inside the venture sees your workflows, what they can do with what they learn, and whether the entity building around your most sensitive processes is structurally aligned with your interests or OpenAI's. Justin Greis, CEO of Acceligence, framed the same worry plainly: "This is a brand new organization with unprecedented levels of access. Who has access to whatever the teams learn?" A new entity with 19 backers and only 15 publicly named is not, on its face, an easy thing to govern.
A Privileged Third Party With Delivery Authority
The structural novelty here is that the FDE is not a contractor you supervise at arm's length. The model makes the FDE a privileged third party with embedded delivery authority, someone building production systems inside your environment rather than handing you specifications to implement. That is a different risk class than hiring a systems integrator, because the builder, the model provider and the deployment entity are now effectively the same supply chain.
Ishraq Khan, CEO of Kodezi, drew the comparison directly: FDE teams "are not like normal software vendors or consultants," and the open question is "who controls the people building around our most sensitive workflows?" His answer is the contract. "Enterprises will need very clear contractual boundaries around data access, model training, employee access," he said. We would add that those boundaries have to be enforceable and auditable, not aspirational. A clause that says your data will not train OpenAI's models is only as good as your ability to verify it.
Why CIOs Cannot Just Say No
The temptation is to treat all of this as a reason to keep OpenAI at arm's length and stick with integrators you already trust. We think that is the wrong reflex. The deployment gap is genuine, the FDE model demonstrably closes it, and the competitive pressure to scale AI from pilot to production is not going to relent because the governance is awkward. Anthropic and others are expanding their own services pushes, and the market is converging on the idea that whoever owns the deployment layer owns the relationship.
So the task for CIOs is not avoidance but governance design. That means treating an FDE engagement like the privileged access it is: scoped data permissions, segmented environments, explicit rules on what learnings can leave the building, and named accountability for the work, not just the equity. The enterprises that win with this model will be the ones that let OpenAI's engineers deep enough to be useful while keeping a hand firmly on who governs the work. Frank Dickson of IDC called the venture logical while flagging exactly this complexity. Logical and complex is precisely the combination that rewards the CIOs who plan for it and punishes the ones who sign first and read later.
