OpenAI Presence & the Rise of the FDE: Who Should Deploy Your Enterprise AI Agents?

If you're deciding how to get AI agents into production, you now have three paths, not two: hire an AI lab's own delivery team (OpenAI's newly launched Presence), build an in-house team, or bring in an independent, model-neutral implementation partner. The right choice comes down to three questions—how much control and metadata you keep, how much vendor lock-in you can tolerate, and whether you've actually built the evaluation and governance to run agents in production. On 22 July 2026, OpenAI introduced Presence, a managed service for deploying agents. The most telling detail isn't the product—it's that even the model maker decided that shipping agents takes engineers on the ground, not just an API key.

What is OpenAI Presence, and how is it different from just calling the API?

Presence is a managed web service that lets enterprises deploy AI agents, with early use cases in customer support, outbound sales, and higher-stakes internal workflows across voice and chat. What makes it notable is how it's delivered. As The Register reports, "Deployments are led by OpenAI Forward Deployed Engineers and select global systems integrators. Presence is not yet available as a self-serve product."

Read that last line twice. The company that convinced the market an API was all you needed to build with AI has concluded that its most important enterprise agents can't be shipped self-serve. Presence isn't "call an endpoint and go"—it's people, on-site, doing implementation work. That's a categorical difference from consuming a model: you're buying a delivery motion, not a token bucket.

What does a Forward Deployed Engineer actually do, and why do enterprises need one?

A Forward Deployed Engineer (FDE) is an engineer who embeds with the customer to turn a general-purpose model into a working, production-grade system for one company's messy reality—its data, its edge cases, its compliance line. OpenAI now runs this as a formal function; The Register notes its delivery arm is "the OpenAI Deployment Company and Forward Deployed Engineers," first stood up last year.

We won't re-litigate the definition here—if you want the full breakdown of the FDE model and value-based delivery, that's covered in what is an AI FDE and value-based delivery. The point worth making today is narrower and sharper: when the leading AI lab makes FDE-led delivery its enterprise strategy, it's top-tier validation that the last mile of AI—not the model—is where projects live or die. The gap between a demo and a deployment is an engineering and process problem, and it needs someone accountable for closing it.

Why is even OpenAI doing hands-on delivery now?

Because the pure-API business doesn't, by itself, get agents into production—and OpenAI is voting with its acquisitions and its headcount. Per The Register, OpenAI expanded its delivery team after acquiring the consultancy Tomoro in May 2026, and, again, Presence is deliberately not self-serve. You don't buy a consultancy and staff up forward-deployed engineers if the API alone were closing the gap.

The demand-side caution is just as important. The Register cites Gartner's forecast that by 2027, about half of the organizations planning to move customer service to AI will abandon those plans, and quotes analyst Kathy Ross: "it is not a panacea… The human touch remains irreplaceable." Even with money pouring in—SoftBank has committed a reported $60 billion to OpenAI and is exploring Presence for support agents—the honest read is that agents fail without someone doing the unglamorous integration, evaluation, and change-management work. Hands-on delivery isn't a nice-to-have; it's the thing that makes the technology stick.

AI lab, in-house team, or independent partner: how should you choose?

This is the real decision, and it's now three-way, not the classic two. The older "build vs. buy" framing still holds as a foundation—see enterprise AI build vs. buy in 2026—but Presence adds a third column: buying delivery from the model maker itself. Here's how the options compare.

Dimension	AI-lab delivery (e.g. OpenAI Presence)	In-house build	Independent, neutral partner (6AM's model)
Delivery model	Led by the lab's FDEs + select systems integrators; not yet self-serve	Hire and build internally	FDE-style, on-site, value-based delivery
Model neutrality	Tied to that vendor's model and stack—potential lock-in	Neutral, but you build the capability yourself	Cross-model neutral; you keep control and your metadata
Speed & cost	Fast, but "boots-on-the-ground" consulting pricing	Slow; hiring is expensive and trial-and-error is costly	Fast, asset-light, focused on your use case
Evaluation & governance	Relies on the vendor's chosen stack	You build eval and governance from scratch	Eval + governance built in; diagnose first, then deploy
Best fit when	You're already all-in on that vendor's ecosystem	You have a long-term in-house AI strategy and budget	You want speed without being locked to a single vendor

There's no universally correct column. If you've already committed to one lab's ecosystem end to end, its delivery team is the path of least resistance. If AI is core IP and you have the budget and the years, build. But if you want to move quickly and keep your options open, a neutral partner is designed for exactly that trade-off.

What's the risk of betting everything on one AI company?

It's the risk of outsourcing your own judgment. Microsoft CEO Satya Nadella put it bluntly, as reported by TechCrunch: "Any firm that doesn't have this control, I will claim will not remain a firm because you've essentially outsourced your thinking."

His prescription is practical and worth internalizing regardless of which vendor you favor: retain the full metadata from every model call—so you can later fine-tune your own weights or move to open models—put an AI gateway layer between your prompts and any single model, and don't lean on a lab's built-in coding harness as your only path. All of that is easier to do when your implementation partner has no incentive to route you into one stack. Neutrality isn't a slogan here; it's the mechanism that keeps the exit door open. That's precisely where an independent, value-based partner differs from a delivery team whose job is to deepen your dependence on its own model.

What to do before you deploy: evaluation and governance decide the outcome

Before you pick a delivery model at all, get honest about whether you can measure and govern what you ship—because the data says that capability is the real predictor of success. Databricks' 2026 State of AI Agents, drawn from more than 20,000 organizations including over 60% of the Fortune 500, found that companies using evaluation tooling get nearly 6× more AI projects into production, and those with AI governance get over 12× more. The bottleneck isn't model access—it's the discipline around the model.

This is why a serious engagement starts with a diagnosis, not a demo. Before writing a line of production code, map your use cases, data readiness, and governance gaps so you deploy the things that will actually hold up. If you want a structured starting point, 6AM's free AI diagnosis is built to surface exactly those gaps before you commit to a delivery path.

The bottom line

Presence is a signal, not just a product: the last mile of enterprise AI is delivery, and even OpenAI now sends engineers to do it. Your job is to choose whose engineers—and on whose terms. Weigh control, lock-in, and your own evaluation and governance maturity honestly, and remember that speed and neutrality don't have to be a trade-off. As an independent, value-based implementation partner, 6AM stays model-neutral and delivers against outcomes rather than lock-in—and the first step costs you nothing but an hour of honesty. Start with a free AI diagnosis.