FDE AI WEEKLY ANALYSIS

When AI vendors sell the engineer, buy the handover

Forward deployed engineering became a product line in 2026, and contracts have yet to catch up

Published: English

Our view

AI vendors are now selling forward deployed engineers as part of the product, which means buyers are paying for a time-limited transfer of capability rather than a system that someone will look after forever. Unless the contract defines handover, ownership of the agent's working assets and the terms of exit, a smooth deployment can quietly turn into lock-in. Japan's trade ministry is now collecting real AI contract problems to inform its guidelines, so contract questions like these may soon get official attention.

The engineer is now a line item

On 6 October CTERA, a data management company, launched a service called Forward Deployed Engineering that places its own engineers inside customer environments to put specific AI use cases into production on the customer's file data. A forward deployed engineer (FDE) is an engineer who works on site with a client to make a product function in that client's real systems. The title now appears in the service catalogue of a data management vendor.

The pattern is not isolated. On 15 July Anthropic, Blackstone and Hellman & Friedman introduced Ode with Anthropic, a services firm that pairs Anthropic's models with experienced AI engineers and operators and aims at mid-size companies. Ode is built on Fractional AI, acquired in May, and its operational core combines that team with engineers from Anthropic. AI News reported that OpenAI's Presence, announced on 22 July, is sold as a project, with deployments led by OpenAI's forward deployed engineers and selected systems integrators, and that access depends on OpenAI's delivery capacity.

FDE AI reads this as a change in how AI is sold, not a staffing fad. Buyers may still be piloting, but on the selling side the engineer-included deal has spread from a services firm co-launched by a model lab and investors to a data management vendor within one quarter. That makes it likely that a buyer's quote blends licence and labour into one line, and if the contract follows a licence-centred software template, the people part may shrink to day rates with nowhere to put handover.

CTERA's chief executive framed the pitch plainly: enterprises are not short of AI tools, they are short of time and specialised skills. That is a fair description of the buyer's problem; it is also a description of a dependency that vendors are now paid to fill.

A handover phase is not a handover clause

CTERA describes each engagement as four phases: discovery and data grounding, a first production use case, scaling, and finally handover so the customer's team can run the work alone. Engagements can be priced by time or as fixed scope with milestone-based acceptance criteria.

Credit where it is due: handover is on the diagram. A diagram in a launch release is a weak basis for a claim, though, and if the signed order form lists only licence fees and day rates, the buyer has little to enforce. A handover defined only as a meeting, a slide deck or a document transfer never tests whether the client can actually operate the agent.

FDE AI reads the handover phase as the point where buyers are most exposed. It comes last, so it is the first budget line to be squeezed when the go-live slips. And if "able to run the work alone" is never defined in advance, the vendor ends up marking its own homework at acceptance.

Agents make all of this harder. An agent is an AI system that carries out multi-step tasks with some autonomy, calling tools and other systems along the way. Accepting a conventional system meant receiving code and documentation. What makes an agent useful sits elsewhere: in the prompts and policies that shape its behaviour, the tool permissions it holds, the evaluation sets used to test it and the log of every adjustment made after launch. If those assets live in the vendor's workspace, the customer inherits a system that works but that nobody on its own payroll can explain.

A rule-maker has started to look

Japan offers the clearest signal. On 14 September its Ministry of Economy, Trade and Industry (METI) began collecting real cases of contract problems and on-site challenges in AI development, provision and use, with submissions closing on 30 October. The ministry will use them to consider revising its AI and data contract guidelines, with an expert panel compiling a report.

FDE AI reads the call for cases as an early draft of the questions procurement teams elsewhere are likely to face. Who is accountable after deployment, who holds the evidence of how the agent was tuned, and what a customer may take with it when the vendor leaves: these are contract questions before they are regulatory ones. Buyers outside Japan do not need to wait for local guidance to start asking them.

The method matters as much as the timing. METI will turn anonymised cases from people doing the work into hypothetical use cases and put them before a panel that includes legal practitioners. That makes the outcome likely to reflect where deals actually break, and FDE AI expects engineer-included agent deployments to be one of those places.

Assumptions and risks

This view rests on two assumptions. First, that the engineer-included model keeps spreading across enterprise software rather than staying with a few model labs. Second, that a large share of an agent's value really does sit in configuration, permissions and evaluation assets rather than in the underlying model.

The strongest counter-argument is cost. For many mid-size firms, relying long term on one capable vendor is cheaper than building an internal team, and strict handover terms may raise the price. On budget alone that may hold. FDE AI keeps its view because even a buyer that chooses permanent outsourcing needs to recover its assets if it switches vendor or the vendor exits. The value of a handover clause is the option it preserves, not a commitment to do the work in house.

Supply is the other risk. If FDE supply tightens, vendors may ration engineers towards their largest accounts. Smaller buyers would then struggle to buy the on-site engineer at all, and the handover problem would arrive later rather than disappear.

Impact on enterprises

Two groups are most affected: mid-size companies evaluating AI agents that plan to buy implementation services from the same vendor that sells the software, and larger firms whose business units run agent pilots through outside partners. The impact lands in three places. Procurement loses the ability to compare prices when licence and labour arrive as one figure. The organisation needs a named internal owner from day one. Compliance teams may be asked to show how a deployed system is monitored and changed, and they need the configuration history and evaluation results in their own hands rather than on request from a supplier. Legal teams should move the people terms from an annex into the main agreement.

Impact on FDE professionals

An FDE will increasingly be judged by what still works after they have gone. Engineers who can write a usable handover pack, design evaluation sets the client can run unaided and package permissions and prompt policies as transferable assets will be in higher demand. For engineers considering a move into the role, fluency in client-side data governance and acceptance testing will age better than expertise in any single model. A new counterpart role is also emerging inside client organisations: the internal engineer who receives the vendor's work. These are shifts in skill demand, not guarantees of pay, placement or a job offer.

Recommended actions

Enterprises

  • When a deal bundles agent software with vendor engineers, procurement and legal should, before signature, make handover a main-contract obligation and define acceptance as the client team completing one configuration change and one evaluation run without vendor help.
  • Delivery leads should name an internal owner before the first use case goes live and secure in the contract either ownership of, or a lasting licence to, the prompts, tool permission lists, evaluation sets and tuning logs, kept in client-controlled accounts.
  • Companies already under contract should, ahead of the next renewal or expansion, audit where those assets are held today and make any gaps a condition of renewal, revoking vendor access when its engineers leave.

FDEs and engineers

  • FDEs starting a new engagement should put handover acceptance criteria in writing at the kick-off meeting and deliver an evaluation set the client can run at every milestone.
  • Engineers aiming for FDE roles should volunteer for handover documentation and permission audits on their next project, building work they can show.
  • Engineers nominated as the client's internal owner should complete at least one configuration change and rollback independently before the vendor leaves and before signing acceptance.

What to watch

  • METI's report or a draft revision of its contract guidelines. If it addresses capability transfer from on-site engineers or ownership of agent configuration assets, handover terms are on their way to becoming standard practice and this view strengthens; if it covers only data and output rights, buyers will have to negotiate these terms themselves.
  • Vendor service terms for new FDE offerings. If more vendors list handover as an acceptance milestone, the market is correcting itself and the urgency of this view falls; if new offerings are built around open-ended subscriptions for on-site staff, dependency risk is rising.
  • Access to vendor engineers. If vendors begin publishing waiting lists or minimum contract sizes for engineer-led deployments, the supply risk above is materialising and smaller buyers should plan for internal capability earlier.

Sources

This article sets out FDE AI's analysis and views. It does not constitute professional investment, legal, accounting, tax, labor or hiring advice. Disclaimer

← All analysis

When AI vendors sell the engineer, buy the handover｜FDE AI