Devoteam is expanding a consulting model that places engineers directly inside client business units, concentrating its AI work on the integration, data, security, and process changes that appear once an experiment has to become an operating system.

Paris based Devoteam says it is on track to field more than 300 forward deployed engineers during 2026, supported by programmes with AWS, Google Cloud, and Microsoft. Its own training scheme will add another route into the practice.

Forward deployed engineers work alongside the employees responsible for the process being changed rather than delivering a system entirely from a separate technology team. Devoteam also expects AI agents to form part of those projects, with embedded teams handling discovery, integration, deployment, and production operation.

Generative AI has made demonstrations comparatively easy to build, but production use exposes dependencies on data quality, identity, security, evaluation, cost controls, existing software, and the detailed exceptions within business processes. Much of that work cannot be solved by selecting a better model.

Engineering is moving closer to operations

An engineer working with the people who run a process can see the informal workarounds and system constraints that disappear when requirements pass through several organisational layers. That proximity is particularly useful for agents, which may need permission to retrieve information from several systems and perform actions rather than simply generate text.

An agent attached to finance, customer service, supply chain, legal, or internal support work can touch identity systems, databases, APIs, and approval flows. Building such a system therefore resembles systems integration and process redesign as much as conventional application development.

Devoteam’s model also alters the commercial shape of consulting. A traditional project can be defined around implementation and handover, whereas embedded engineering keeps the supplier closer to the continuing operation of the technology and can blur the line between consultancy, software development, systems integration, and managed service.

The company says its engineers will themselves use AI agents as part of delivery. It has not published comparative figures showing how much that reduces cost or deployment time, leaving the productivity case to be demonstrated across live projects rather than inferred from the presence of automation.

Cloud providers want implementation capacity

Devoteam’s expansion is closely linked to its largest platform partners. The company has a five year strategic collaboration agreement with AWS for forward deployed work across EMEA, participates in Google Cloud’s FDE enablement programme, and has been selected as a Frontier Partner in Microsoft’s corresponding ecosystem.

A Google Cloud cohort has already trained on building, securing, evaluating, and operating production AI agents using Gemini Enterprise, covering areas such as multi-agent architecture, identity, guardrails, integration, and cost optimisation. Devoteam says the initial group is part of a wider certification target.

Cloud providers have a commercial interest in increasing this implementation capacity because a production AI system consumes compute, model services, databases, observability, security, and other infrastructure over time. A pilot that never progresses beyond a controlled test generates much less durable consumption.

The relationship can benefit customers by giving engineers deep familiarity with the platforms they are deploying, but it also creates a procurement consideration. Partner certification and training naturally influence which technology stacks can be implemented quickly, even where a consultancy works across several providers.

Customers therefore need to examine whether the embedded team is solving the operating problem or primarily accelerating consumption of one vendor’s services. Those aims can overlap, although data architecture, portability, and exit planning become more important as an AI workflow grows dependent on one provider’s models, identity services, agent framework, and data layer.

Devoteam employs around 11,000 specialists across more than 25 EMEA countries, giving it an existing workforce through which the model can spread. The harder test comes after deployment, when clients have to operate and modify the systems without keeping a permanent external engineering team attached to every process.

If forward deployed engineering is to close the gap between AI pilots and production, its output cannot simply be another long running consultancy dependency. The stronger result will be software, governance, and internal capability that continue to work when the embedded engineers eventually move on.