OpenAI has finally acknowledged the obvious: in their current form, custom GPTs remain little more than corporate toys, incapable of surviving a collision with real-world business operations. With the launch of Presence, Sam Altman’s company is attempting to pull AI agents out of the internal prototype sandbox and embed them into customer service and complex workflows. Until now, agentic functions in Workspace were confined to internal perimeters, but Presence targets the messy reality of the enterprise segment, where model hallucinations and the resistance of legacy IT infrastructure turn any deployment into a nightmare.
The biggest surprise is the effective abandonment of the idea that AI can be implemented with the click of a button. OpenAI is deploying a landing party of Forward Deployed Engineers (FDEs). These specialists will manually refine client systems, selecting viable use cases and building operational guides. This move looks like a tactical surrender to reality: autonomous agents are still not reliable enough to operate without close engineering supervision. Instead of pure software, OpenAI is now offering a hybrid of neural networks and expensive consulting to somehow domesticate the technology in high-stakes environments.
Infrastructure standards through manual labor
For now, Presence remains a closed club for select enterprise clients. Meanwhile, OpenAI maintains an eloquent silence regarding the legal side of the matter: details on compliance with the strict EU AI Act or regional regulations remain murky. It is clear the company is trying to cement its status as the primary infrastructure provider for business agents, but it is doing so through intensive technical support rather than elegant code.
In our view, this transforms OpenAI from an IT giant into something resembling Palantir, where product success depends directly on how many engineers are "embedded" to solve a specific customer task.
Key points
OpenAI is deploying implementation engineers to manually bridge the reliability gap in its systems.
This is a logical, albeit expensive, way to scale in a world where neural networks alone are no longer enough to automate real-money transactions.
The bottom line
The shift toward a bespoke support model highlights the immaturity of current autonomous solutions. The market leader admits that for actual enterprise-grade work, AI still requires human crutches and deep manual tuning to meet the needs of each specific client.