On September 29, Gartner published a prediction that summarizes the next two years of the enterprise AI market: by 2028, 70 percent of enterprises will abandon agentic AI systems built by vendor forward-deployed engineering (FDE) teams — trapped by soaring costs and unable to evolve the systems on their own.

Seventy percent is not an implementation-detail warning; it's a business-model warning. In this guide, we explain why those organizations will walk away — and how to design ownership so that yours isn't among them.

The Anatomy of Walking Away: Why the Model Attracts, Why It Risks

The forward-deployed engineering model works like this: the vendor embeds its own engineers in your organization; they build the system in your environment, they tune it, they keep it alive. The early results impress; the speed arrives without the effort of building an internal team.

The risk hides inside that very speed. The system runs, but the knowledge belongs not to the organization — it lives in the vendor's engineers' heads. When a process changes, a new integration is needed, or an AI agent's behavior must be adjusted, you turn to them again. Every change is an invoice; every delay is evidence of dependency.

This is the breaking point Gartner foresees: cost rises, the ability to evolve does not. At some point, the organization is left with a system that works but that it cannot touch. What gets abandoned in 2028 won't really be the technology — it will be deployments whose ownership was never transferred.

Gartner's companion message completes the picture: success with AI agents requires reliability before autonomy. And reliability cannot be rented from outside; you cannot keep reliable a system you don't understand.

What Is Ownership? Four Concrete Components

"Ownership" sounds abstract; it isn't. It has four deliverable components:

1. Readable architecture and documentation. Which AI agent does what, which systems does it access with which authority, where do exceptions flow? This knowledge must live in current documents the organization can access — and in the architecture itself — not in an engineer's head.

2. Transfer to the internal team. Knowledge transfer must be a work package planned from the start, not a closing meeting on the project calendar: the organization's team grows with the system. There is no handover day; there is a handover process.

3. Standard connection points. The system should be built on the industry's shared standards, not on vendor-specific closed couplings. In an architecture built on standards, parts can be replaced; even if the vendor changes, the structure stands.

4. The organization's own path to evolve. The critical question: when a change is needed tomorrow, who takes the first step, and with what knowledge? If the answer sits entirely outside the organization, what you have is not a system you own but a service you rent — and the rent rises every year.

Five Questions for the Procurement Table

The checklist we recommend to teams evaluating vendors for an AI agent project:

1. When the project ends, in what format does the documentation, process definitions, and agent configurations remain — and in whose ownership?

2. Is knowledge transfer a separate work package in the proposal? How many hours, to whose team?

3. Which open standards does the system support, and what is the list of vendor-specific components?

4. After the contract ends, under which support model will our internal team be able to make a mid-sized change (a new exception rule, a new integration)?

5. What is the three-year projection of annual maintenance and change costs, and what drives the increases?

The answers to these questions don't show up in the product demo. They show up in the contract.

Where We Stand

As a team that has been building process automation and AI agent architectures for years, our model is clear: the architecture, documentation, and process knowledge of every system we build belongs to our client, and the internal team is involved from day one. We prefer to earn our clients' continued business not through necessity, but through the desire to build the next process together.

Because the picture is clear: the client held by dependency leaves at the first opportunity — Gartner's 70 percent is the proof. The client strengthened by ownership comes back to grow. Long-term business is built on trust, not on locks.

Conclusion

In AI agent projects, what is really delivered is not the system — it is ownership. Most organizations will deploy this technology within the next two years. The way to stay out of 2028's 70 percent walk-away wave is to write a single question into the contract today:

When this project ends, what remains in our hands?

Sources: Gartner, "70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028" (September 29, 2026) and "AI Agent Success Requires Reliability Before Autonomy" (October 2026); October 2026 industry analyses.