Forward-deployed engineering has become one of the most talked-about delivery models in enterprise AI. Put strong engineers inside the customer's environment, build against real systems and real data, and get agents into production fast. It works. The harder question is what happens after the engineers leave.
On September 29, Gartner put a number on that question. It predicts 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.' For banks, telecoms, hospitals and energy companies, where an abandoned system is not just wasted budget but an operational and compliance risk, that warning deserves a careful read.
We run forward-deployed teams ourselves, so we read it closely. Our view is that Gartner is describing a real failure mode, and that it is avoidable if handover is treated as a design requirement from day one rather than a phase at the end.
Why vendor-built AI gets abandoned
Gartner's analysis points to two causes: cost and dependency. In Channel Dive's coverage, Gartner estimates that FDE consulting fees alone could top $200,000 per quarter for a single use case. That is sustainable for a pilot. It is not sustainable across dozens of workflows.
Dependency is the deeper issue. As Gartner analyst Mukul Saha told IT Brew, the worry is 'clients not having the right internal capability to take over and run' when the vendor's term ends. An agent that only its builders understand cannot be safely changed when a regulation, a product or a risk appetite changes. So it gets frozen, then bypassed, then switched off.
Gartner also warns about 'FDE washing': traditional consulting relabelled as something more strategic, and predicts that fewer than 20% of FDE engagements will turn recurring customer needs into real product capabilities through 2028. In other words, a lot of what gets built is bespoke work that nobody but the vendor can maintain.
Ownership is an architecture decision
It is tempting to treat handover as a training problem: run workshops, write documentation, hand over the repository. Those things matter, but they come too late if the system was never built to be owned by someone else.
A system is ownable when the client's team can answer three questions without calling the vendor:
The rest of the market is converging on the same answers. In the past few weeks, Microsoft introduced dynamic permissions for agents, where agents start with minimal access granted per task and high-impact actions require human review. Google announced an Agent Registry and per-agent identity, a central source of truth for approved agents, tools and models, with OpenTelemetry tracing for every model call and tool invocation. IBM expanded watsonx.governance to tie observability metrics to governance controls. The direction is clear: permissions, identity, audit and evaluation are becoming standard infrastructure, not custom work.
That matters for ownership. When permissions, audit trails and evaluations live in a shared operating layer instead of inside each bespoke agent, the client inherits a platform with rules it can read, not a collection of one-off scripts.
A practical handover checklist for regulated enterprises
Gartner frames successful FDE engagements in three phases: before signing, during the engagement, and at exit. Here is how we translate that into concrete requirements a CIO or COO can put in the contract and the project plan.
Before signing
During the engagement
At exit
What this means for Intelligent's approach
Our model is built around the same idea: local engineering teams build agents and systems on top of a governed operating layer, in the client's own environment, and then move the client from vendor-led to client-owned operation in defined phases. Permissions, audit trails and evaluations are built once in the platform and reused, so what we hand over is something your team can read, test and change.
Forward-deployed engineering is a powerful way to get AI into production quickly. It only creates lasting value if the organization can run what was built after the vendor steps back. The 70% figure is a forecast, not a destiny. The enterprises that avoid it will be the ones that ask about handover in the first meeting, not the last.
Planning an agentic AI program and want to make sure your team owns it from day one? Book a meeting with our team at intell.web.app.
Sources & Industry References
- Gartner: Gartner Predicts 70% of Enterprises Will Abandon Agentic AI Built by Vendor Forward-Deployed Engineering by 2028 (Sep 29, 2026)
- The Register: 7 in 10 enterprises expected to abandon vendor-built agentic AI by 2028 (Sep 30, 2026)
- Channel Dive: Are forward deployed engineers flirting with failure?
- IT Brew: Most enterprises will abandon vendor FDE-built AI by 2028: report
- SecurityWeek: Microsoft Targets Rogue AI Agents With Dynamic Permissions, Risk Scoring
- Google Cloud Blog: Introducing Agent Compute Engine and expanding agent governance (Oct 22, 2026)
- IBM: New in IBM watsonx.governance (Oct 14, 2026)
