Forward-Deployed AI Now Has to Prove Independence

The 2026 land grab for forward-deployed engineers (FDEs) — vendor engineers embedded inside customers — has produced staggering numbers. AWS committed $1 billion to its own FDE unit, seeded with "thousands" of engineers who deploy in pods of roughly five or six per customer S3. OpenAI launched a deployment business backed by more than $4 billion, Anthropic is putting $100 million into training 10,000 FDEs by end of 2027, and Accenture and Google Cloud are building a 1,000-person FDE workforce S2. US FDE job postings grew from under 200 in 2023 to more than 7,300 in the first nine months of 2026, with median advertised salaries up 40 percent S2.

Then on September 29, Gartner gave the boom a deadline: by 2028, 70 percent of enterprises will abandon agentic AI built by vendor forward-deployed engineering, "trapped by soaring costs and unable to evolve it on their own" S1.

The forecast measures dependency, not model quality

Read the announcement closely: the failure mode is not accuracy or hallucination. It is ownership. "FDE engagements often fail structurally before they fail technically" S1. Gartner's concern is that customers may see faster early progress but fail to build internal capability, ending up paying premium rates for work that — once the scope is clear — a traditional services or partner model could potentially deliver more cost-effectively S1.

Gartner adds a vendor-side warning: through 2028, fewer than 20 percent of FDE engagements will turn recurring customer needs into capabilities in the vendor's core product, and it flags "FDE washing," conventional consulting rebranded under the FDE label S1.

The deliverable is the exit

Gartner structures a genuine FDE engagement in three phases — before signing, during delivery, and "exit and prove independence" S1. In practice: deliverables, knowledge transfer, IP rights and exit responsibilities set at signing; FDEs embedded alongside internal domain experts during the build; an exit plan executed rather than quietly extended because the internal team feels unready. Success is measured not by implementation completion but by whether the enterprise can manage, optimize, and scale the system on its own S1.

Vendors publicly agree. Per CNBC, AWS's FDEs work alongside customer engineers so those teams can eventually maintain and expand the systems themselves S2, and AWS has said its pods aim to leave behind self-sufficient teams in a matter of weeks S3. Asa Kalavade, AWS's vice president of forward-deployed engineering, framed the spend the same way: "We are investing $1 billion now because customers need this hands-on help turning their AI investments into results at speed" S2. The gap between that intent and Gartner's forecast is where delivery discipline lives.

Three checks before you sign

• Is the problem FDE-shaped? Gartner reserves the model for problems requiring deep product expertise, rapid adaptation, or close integration between vendor technology and the customer's operating environment S1. If a standard services engagement would do, the premium buys dependency risk.
• Who owns day 100? If your engineers cannot make a routine change without the vendor, you have bought dependency, not delivery. Knowledge transfer and exit terms belong in the contract, not the handshake.
• Where do repeats go? Ask which recurring needs become product features versus a billable custom branch only the embedded team understands.

The best evidence so far from FDE projects is productivity improvement, not independently verified financial returns S2; as Caylent's CEO put it, "There is no compression algorithm for experience" S2. Budget for the unglamorous handover work, pair your engineers with the visiting pods, and treat the exit exam as the acceptance test. The best FDE engagements make themselves unnecessary.