OpenAI Presence: Why Model Vendors Are Switching to Forward Deployed Engineers for AI Agents
OpenAI is delivering its new Presence product through Forward Deployed Engineers rather than a self-serve API — a signal that enterprise AI-agent delivery is moving from "sell an API" to "embed on-site." We unpack the shift and give you a four-factor framework for choosing between a model vendor's FDE team and an independent implementation partner.
TL;DR: The forward deployed engineer is becoming a default way enterprises get AI agents into production. OpenAI's new product, Presence, is delivered not as another self-serve API but through the company's own Forward Deployed Engineers (FDEs) and select systems integrators, according to a report in The Register. When even the model vendor ships a product this way, the practical question for buyers shifts from "which API" to "which delivery partner." Weigh that choice on four factors — lock-in risk, neutrality, long-term partnership, and cost — and start by mapping where an agent would actually create value with a free AI readiness diagnosis.
What is OpenAI Presence, and how is it different from a self-serve API?
Presence is a web service for deploying AI agents into live voice and chat scenarios — customer service, outbound sales, and high-risk internal workflows — with a control plane that includes an agent editor, a playground, tools for actions like refunds, order status and account deletion, and policy-based governance controls. The distinguishing detail is in how you get it. As The Register reports, "rather than releasing another self-service API, OpenAI is making Presence available through its consulting arm, the OpenAI Deployment Company and Forward Deployed Engineers" — a delivery workforce that debuted last year and was strengthened by OpenAI's acquisition of the consultancy Tomoro in May. The report is blunt: "Presence is not yet available as a self-serve product," and deployments "are led by OpenAI Forward Deployed Engineers and select global systems integrators."
That is a meaningful reversal. A forward deployed engineer is an embedded engineer who works inside the customer's environment to make software actually deliver an outcome — the model popularized by Palantir and now adopted by the frontier labs themselves. Pricing follows the same logic: The Register notes Presence is scoped per customer use case, and SoftBank — which has committed $60 billion to OpenAI — is piloting it for Japanese-language customer service. When the product is priced per engagement and installed by engineers, you are buying a consulting relationship, not an API key.
Why are model vendors shifting to on-site delivery instead of selling APIs?
Three forces are pushing agent delivery toward the embedded model.
Agents are a systems problem, not an inference problem. In a piece for MIT Technology Review, Intel argues that "an agent is a goal-driven automated enterprise workflow process… enterprise agents are therefore not just an inference problem; they are a systems problem." Drawing on thousands of agentic-workload experiments, Intel's team stresses that enterprise value depends on the whole system — orchestration, data access, tool execution, latency, governance, and observability — not on the model in isolation. A raw API hands you the inference and leaves every one of those hard parts to you, which is precisely why enterprise AI deployment is hard.
The data says AI is augmenting work, not replacing it. Google Research's "AI & Economy ATLAS" study, covered by Ars Technica, analyzed 15 million anonymized interactions and "did not find evidence… to support the claims that AI is about to cause massive automation and displacement of white-collar work." Only 21% of work tasks were classified as "Gemini tasks," 29% of occupations had no task cross a significant usage threshold, and just 3% of occupations used Gemini for at least 75% of their relevant tasks; 86% of interactions were cognitive tasks concentrated in drafting and information retrieval. The study's conclusion — that AI is "currently serving primarily as a complement" to existing work — helps explain why a human-led, embedded rollout tends to outperform hand-off-and-hope self-service.
High-risk self-service is already retreating. The Register cites a Gartner forecast that by 2027, half of organizations planning to move customer service to AI will abandon those plans, and quotes analyst Kathy Ross: "AI… is not a panacea. The human touch remains irreplaceable in many interactions." Agents that take real actions fail differently from chatbots that only talk — a reliability gap we cover in why AI agents fail in production. On-site delivery is one way vendors appear to be responding to that gap.
Model-vendor FDE vs an independent implementation partner: how should you choose?
Both models put engineers on the ground; the difference lies in their structural incentives. Treat it as a build-vs-buy decision one level up — not "build or buy the agent" but "whose delivery model do we buy into" — and score it on four factors.
| Factor | Model-vendor FDE (e.g. OpenAI Presence) | Independent implementation partner (e.g. 6AM) |
|---|---|---|
| Lock-in risk | May increase dependency on a single model/cloud ecosystem, raising switching cost | Lower — model-neutral selection per scenario; multi-model by design |
| Neutrality | Structurally aligned with the vendor's own model and product roadmap | Vendor-agnostic by design; oriented to business ROI |
| Long-term partnership | Often project-based delivery; ongoing iteration may track the vendor's roadmap | Long-term partnership; continuous tuning and governance as the business evolves |
| Cost | Scoped per use case, potentially at a frontier-vendor premium | Priced to delivered value, focused on observable efficiency gains |
The verifiable anchors for the vendor column come straight from The Register: delivery through OpenAI's consulting arm, no self-serve option, and pricing scoped per use case. Those facts can tie a Presence engagement more closely to one vendor's ecosystem — not a criticism of the product, but a structural consequence of buying delivery from the party that also sells the model. The rest of the table describes structural tendencies you should test against your own contract terms, not guarantees.
Where does an independent forward-deployed partner add value?
An independent FDE partner's strength is the neutrality that is harder for a model vendor to offer: choosing the right model per scenario, and measuring agents across vendors rather than inside one.
Measurement is where this gets concrete. Intel's MIT Technology Review piece recommends planning capacity by agent density (agents per vCPU) rather than agent count, and describes extending the open-source Terminal-Bench with deterministic record-replay to separate an agent's performance from the LLM's run-to-run variance. That kind of vendor-neutral, reproducible evaluation is exactly what an independent partner builds into a deployment — the discipline behind production AI-agent reliability and observability and continuous agent observability, audit, and evaluation in production. Because a model vendor and an independent partner have different incentives here, it is worth writing cross-model evaluation into the deployment contract rather than assuming it.
The other half of the value is staying past go-live. Enterprise adoption tends to succeed or fail on organizational fit — cost, permissions, stability, and security adapted to how the business actually runs — as much as on model quality. An embedded partner tunes those over time, which is why 6AM's delivery model pairs on-site engineering with long-term operation rather than a hand-off at launch.
If you are weighing your options, three next steps: run a free AI readiness diagnosis to find where an agent would create measurable value, browse our courses to level up your team, or check the FAQ for common deployment questions.
Frequently asked questions
Is OpenAI Presence a self-serve product?
No. As The Register reports, Presence "is not yet available as a self-serve product" — it is delivered through OpenAI's Forward Deployed Engineers and select global systems integrators, with pricing scoped per customer use case.
What is the biggest difference between using a model vendor's FDE team and an independent implementation partner?
Lock-in risk and neutrality. A model vendor's delivery is structurally aligned with that vendor's own model and products, which can increase single-ecosystem dependency. An independent partner selects models per scenario, which helps keep switching costs down and evaluation vendor-neutral.
Do small and mid-sized businesses also need "embedded" AI deployment?
Yes. The hard part of adoption is often organizational — adapting cost, permissions, stability, and security to how the business runs — not the model itself. Smaller teams have less slack to absorb a failed self-service rollout, so a guided, embedded approach that starts from a concrete use case tends to pay off faster.
