The Invoice for Making AI Real: Gartner Says 70% of Enterprises Will Abandon Vendor-Built Agentic AI by 2028

For two years, the enterprise AI industry has had an open secret: the agents don’t deploy themselves. When frontier models proved unable to survive contact with real corporate workflows — legacy systems, fragile integrations, compliance obligations, undocumented workarounds — vendors responded by shipping humans alongside the software. These are forward-deployed engineers (FDEs), vendor engineers embedded directly inside customer organizations to build, integrate, and babysit agentic AI systems.

On September 29, 2026, Gartner put a hard number on where this model ends. In a press release titled around its headline prediction, the analyst firm forecasts that by 2028, 70% of enterprises will abandon agentic AI built by vendor forward-deployed engineering, “trapped by soaring costs and unable to evolve it on their own.” And in a second, quieter prediction, Gartner adds that through 2028, fewer than 20% of FDE engagements will turn recurring customer needs into capabilities in the vendor’s core product — exposing what the firm calls a growing risk of “FDE washing,” where ordinary consulting services are re-marketed under the more strategic-sounding FDE label.

Why this prediction lands hard

The FDE model is not a fringe experiment — it is currently the deployment backbone of the agentic AI industry. Palantir pioneered the playbook; OpenAI, Anthropic, Google, and AWS have all scaled variants of it. Just three weeks ago, Google Cloud and Accenture announced a jointly funded unit with a planned 1,000-person FDE workforce to move enterprise AI past the pilot stage. Earlier this year, financial technology vendor FIS made a point of highlighting that Anthropic’s FDEs were embedded with its team to co-design a Financial Crimes AI Agent deployed at Bank of Montreal and Amalgamated Bank.

In other words, the industry’s answer to “why don’t your agents work in production?” has been to put the vendor’s engineers in the building. Gartner’s prediction says that answer has a shelf life of roughly two years.

The core mechanism is a dependency trap. Enterprises pay premium consulting rates for teams whose knowledge never fully transfers. When the engagement ends — or the invoices pile up — the customer is left holding a system it cannot operate, monitor, challenge, or safely modify without the vendor. As Gartner senior director analyst Alex Coqueiro framed it in earlier research: “Flat FDE effort across successive deployments is the signal that an engagement has produced a dependency, not a capability. When effort does not decrease as use cases mature, the organization is paying consulting rates for operations it should own.”

The three-phase playbook Gartner prescribes

The September 29 release, built around research from Sr Director Analyst Mukul Saha, argues that FDE engagements “often fail structurally before they fail technically.” The outcome is decided by choices made before contracting, during delivery, and at transition. Gartner breaks its guidance into three phases:

Phase 1 — Before signing. Use FDE only for problems that genuinely require deep product expertise, rapid adaptation, or tight integration between vendor technology and the customer’s operating environment. Appoint an executive sponsor accountable for business outcomes, not just the implementation budget. Hammer out contractual requirements beyond procurement’s usual scope: deliverables, knowledge transfer, intellectual property rights, and transition or exit responsibilities — from day one. As Saha puts it: “FDE success starts with getting the engagement structure right, from scope and incentives to governance, ownership, and exit.”

Phase 2 — During the engagement. Embed the FDEs with internal domain experts, engineers, and end users so critical knowledge actually moves and the solution reflects how work really gets done — not how the org chart says it does. Establish a cadence of iterative business validation that evaluates not only technical performance but the operating model required to scale AI responsibly: decision rights, the balance between agent autonomy and human oversight, and how confidence and exceptions get communicated.

Phase 3 — Exit and prove independence. Execute the exit plan established at the start, rather than extending the engagement indefinitely because internal teams aren’t ready. “Success is measured not by implementation completion, but by the enterprise’s ability to manage, optimize, and scale the technology,” Saha said — including adapting human-AI decision models as business processes and risk profiles evolve.

“FDE washing” and the 20% problem

The second prediction may sting vendors more than the first. Gartner expects that through 2028, less than a fifth of FDE engagements will convert recurring customer needs into shipped product capabilities. That is the entire theoretical justification for premium FDE pricing: the vendor learns from the deployment and productizes what it learns, so the next customer buys software instead of people.

If 80%+ of engagements never make that leap, the FDE label degenerates into branding. “Many providers now use ‘forward deployed’ as a label for implementation, professional services, solution engineering, or AI consulting; some thoughtfully, others because it sounds more strategic,” Saha said. “Some charge premium fees without the delivery depth, program management, or change management maturity to justify them.”

The result for buyers: faster early progress, but no internal capability — and premium rates for work that, once the scope is clear, a traditional services firm or partner could deliver more cheaply and predictably.

A pattern, not a one-off

Today’s forecast extends a now-familiar series of Gartner downgrades to the agentic AI hype cycle:

• June 2025: over 40% of agentic AI projects will be canceled by end-2027, due to escalating costs, unclear business value, or inadequate risk controls.
• April 2026: more than half of enterprises will stop paying for assistive AI (copilots, smart advisors) by 2028, in favor of outcome-focused workflows.
• August 2026: AI inference costs per agentic workflow will increase more than fivefold through 2028.

Read together, the arc is consistent. The technology keeps improving; the economics of deploying it keep getting worse. Agents that work in demos require expensive humans to work in production, and each agentic workflow consumes more inference than the last.

What it means

For CIOs, the actionable test is the one that emerges from the whole report: after the forward-deployed team leaves, can your organization still operate, monitor, challenge, and safely modify the agentic workflow? If the answer is no, you haven’t bought a capability — you’ve rented a dependency with better stationery, and Gartner’s 70% is where that road ends.

For vendors, the message is that the FDE gold rush has a countdown clock. The model works as a bridge; it fails as a business model if the bridge never reaches the other side. The vendors that convert engagement learning into product at better than a 20% clip will own the enterprise agent market after 2028. The rest will discover that their customers did the math.