Articles

What 25 Years Inside Client Businesses Taught Us About Forward Deployed Engineering

November 19, 2026
7 minutes
‌

At a glance

  • Only 33% of organizations say they capture the full value of AI, according to the Xebia Data & AI Monitor 2026.
  • A Forward Deployed Engineer (FDE) operates inside the client’s business, where business, customer, consulting, engineering, and AI meet.
  • For 25 years, Xebia's engineers have worked inside clients’ businesses, solving problems where the work happens.
  • Xebia’s FDE approach is built around the pod: a Domain FDE, a Delivery FDE, and an AI FDE work as one pod on one defined problem, accountable together for an outcome agreed at the start.
  • More than 200 of our professionals have completed Xebia’s FDE capability journey so far.

Most enterprises I speak with have done an AI pilot. Far fewer have one running in production and paying back what it promised. 

There is enough evidence to support this.

In the Xebia Data & AI Monitor 2026, only 33% of respondents say they capture the full value of AI. Agents show the same gap. Of the organizations surveyed, 28% have agents in production, but only 7% have extended them across multiple processes, and 59% are still piloting or exploring. Another 26% name moving data-driven solutions into production as a top barrier. Gartner’s mid-2025 forecast points the same way: more than 40% of agentic AI projects will be canceled by the end of 2027.

When I look at pilots that stall, I don’t think the models are the main problem. Where they get stuck is the way the work is organized around it. Most enterprise delivery still follows a familiar pattern. Someone writes the requirements. Someone else builds against those requirements. Users see the result near the end. That pattern assumes you can describe a system fully before you build it. With AI, that assumption doesn't hold, because you only find out what an agent can do once real people use it on real work. A program built on handoffs tends to learn that after most of the budget has gone.

A Forward Deployed Engineer, or FDE, operates inside the client’s business, where business, customer, consulting, engineering, and AI meet. They frame the problem with the people who own it, build the solution in the client’s own environment, and stay with it through production. What sets an FDE apart is accountability. An FDE is judged by changes in the business: cycle time, cost, decision quality, and revenue. The code matters, but nobody measures an FDE by it.

That’s why I see FDE as an operating model more than a job title. Traditional delivery defines the work before the build. An FDE team learns the work as it builds it. Autonomy comes in steps, and each step needs evidence. The agentic enterprise is one where AI agents perform real steps of the work, within the limits of the business. People work less and supervise more.

Nobody gets there in one move. I think of it as a ladder:

  • AI assists. It gathers data, drafts and summarizes, and people do the work.
  • AI proposes. A person approves each action. 
  • AI acts within limits. Anything above a set threshold goes to a person. 
  • AI runs the step. People monitor and audit it.

The level applies to each step of a process, not to the whole company. Take an invoice that doesn’t match its purchase order. The same exception can be handled at level one in one team and level three in another. 

Buying a better model doesn’t move a step up the ladder. Four things have to be true first: 

  1. The work is mapped, with a baseline for cost and cycle time. 
  2. Evaluations pass on real data, real permissions and real exceptions.
  3. The platform checks every change before it reaches a live system. 
  4. Each agent has monitoring and a way to stop it, and a person signs off on anything that can’t be undone. 

None of that can be done well from a distance. It needs someone sitting next to the people who know the job, and that's where an FDE comes in. 

Six competencies, rarely found in one person.

So, what does a good FDE look like? 

  1. Generalist engineering depth: shipping production code across the stack. 
  2. Problem framing in ambiguity: turning a messy operation into something buildable.
  3. AI and agentic fluency: designing and deploying agents in real operations.
  4. Domain depth: knowing an industry well enough to define what “correct” means. 
  5. Consultative communication: bridging the gap between the engineering team and the boardroom. 
  6. A reuse mindset: turning each engagement into assets the next one builds on. 

People with all six competencies do exist, but they’re rare. They take years to develop, and even then, one person can work on only one client at a time. An enterprise strategy that depends on finding those people will struggle to scale.

Why Xebia chose the pod 

None of this is new to us at Xebia. For 25 years, our engineers have worked inside clients’ businesses, solving problems where the work happens. FDE gives that way of working a name, and our pods give it a structure.  

That experience made one thing clear early. The competencies an FDE needs rarely sit in one person. So, we split the role into three.  

Domain FDE

Knows the business

Delivery FDE

Gets it to production

AI FDE

Makes it intelligent

BRINGS
  • Industry and functional depth
  • Business process and data understanding
  • Regulatory and risk context
  • Consulting and discovery skills
  • Software engineering and architecture
  • Cloud, integration and security
  • Quality, DevOps and platform practices
  • Experience in complex enterprise estates
  • AI/ML and generative AI
  • Agentic systems and data engineering
  • Model routing, agent orchestration, guardrails and evaluation
  • AI-assisted development practices
OWNS
  • Framing the right problem
  • Defining what success looks like
  • Mapping domain workflows and data
  • Validating fit with how the business works
  • Measuring business impact
  • Solution architecture and engineering decisions
  • Integration with existing systems
  • Enterprise standards and non-functionals
  • Productionization and operational readiness
  • Handover and run
  • AI solution design
  • Rapid experimentation and prototyping
  • Model and agent selection and evaluation
  • Responsible AI practices
  • AI-accelerated engineering across the pod

  • The Domain FDE knows the business. This person finds the right problem, defines what success looks like and measures the business impact. 
  • The Delivery FDE gets it to production. This person owns the architecture, the integration with existing systems and the handover to the client’s run team. 
  • The AI FDE makes it intelligent and safe. This person chooses the models and agents and builds the evaluations and guardrails that prove they work. 

The three work as one pod on one defined problem, accountable together for an outcome agreed at the start. They lead a wider team of engineers and AI agents. The problem never changes hands between strategy, engineering and deployment teams. 

Why three? Because these are three distinct responsibilities, and each one is something we can hire for, train and grow. More than 200 of our professionals have completed Xebia’s FDE capability journey so far. That’s how a pod model scales where a search for rare individuals can’t. 

Governance is part of the build from the beginning. And when the pod rolls off, the client’s own teams own the system. 

What changes on a real workflow 

Here’s an illustration, not a client result. 

A claims-triage agent scores well in the lab. In its first week in production, it misses an exception rule that only one senior handler knows. It can’t write to the claims system with a proper audit trail. And it misreads a scanned form it was never tested on. 

A pod catches each of those, because each sits with a different role. The Domain FDE learns the unwritten rule by working alongside the handlers. The Delivery FDE builds the integration and the audit trail. The AI FDE tests the model on real claims, including scanned forms, before every release. 

Three things change when a team works this way: 

  1. The team sees the exceptions that never make it into a requirements document. 
  2. It’s measured on a result the business agreed to before the build. 
  3. What it learns stays with the client after the team leaves. 

Where to start, and where not to 

This model doesn’t fit every problem. It isn’t worth the investment where the value is low, where the work is simple and rule-based, or where no business owner will take responsibility for the result. Where it does fit, start small. Pick one workflow that matters, with an owner in the business who wants it fixed. Put a pod next to the people who do the work, agree the baseline, and move one step up the ladder when the evidence says you’re ready. 

Enterprises that take that first step will learn more in a quarter than many learn from a year of pilots. The models will keep getting better for everyone. The real advantage will go to organizations that put their engineers next to the work. 

That’s been our experience over 25 years of working inside client businesses. We welcome the opportunity to partner with you in transforming your organization with Forward Deployed Engineering. 

Learn more about Xebia.AI/FDE

Frequently Asked Questions

Written by

Anand Sahay

Global CEO

Anand Sahay is the CEO and Executive Director at Xebia, a global AI-first, digital transformation, and engineering partner operating in 17 countries. A seasoned technology strategist, Anand helps Fortune 500 companies modernize at scale through cloud-native solutions, AI/ML integration, and intelligent automation. Under his leadership, Xebia has built industry-leading capabilities in generative AI, data engineering, and scalable digital platforms. Known for his ability to align innovation with business value, Anand partners with C-level leaders to drive enterprise-wide transformation and long-term growth in the AI era.

Contact

Let’s discuss how we can support your journey.

‌
‌
‌
‌
‌
‌
‌
‌
‌