Six years ago I was asked to put together a forward deployed team for an enterprise AI implementation at a Fortune 500 customer. I hired two of the strongest AI engineers I knew, briefed them on the problem, and sent them in.
What they built was technically excellent. It was also unused.
The model worked. The integration worked. The dashboard worked. And the customer's team — the people whose workflow we were supposed to be transforming — politely routed around all of it within four months.
I learned something that quarter that I have spent six years trying to teach the rest of the industry:
The "forward deployed engineer" is the wrong unit of analysis.
The right unit is a forward deployed AI team. And almost no one in the industry is building one.
What I got wrong
The mistake was thinking the problem was engineering. The model was not accurate enough. The integration was clunky. The latency was too high.
Those things were all true. They were also not the reasons the customer didn't adopt.
The actual reasons:
- We had built the wrong thing — not because the technology was bad, but because we had never sat with the user long enough to understand the workflow we were interrupting.
- The system architecture, while functional, made the customer's IT team nervous. They couldn't see how it would survive their next infrastructure migration.
- No one on our team was making the daily judgment calls about what to cut, what to defer, and what to ship to keep momentum.
I had hired engineers and given them a job that needed a designer, an architect, a product person, and engineers — working as a single forward deployed unit.
The model that actually works
The forward deployed AI team has four roles. None of them is optional. Click any role to see what they own and why they matter.
Why design thinking is the connective tissue
Every AI team that has worked under me used some form of design thinking — empathize, define, ideate, prototype, test — adapted for AI deployments.
Why design thinking specifically? Because AI projects have a unique failure mode: the technology works exactly as specified, and the project still fails.
You will not encounter this in a traditional software project. Traditional software fails when the code doesn't compile, the API breaks, or the database falls over. AI projects fail when the model is 94% accurate, the integration is solid, and the user still doesn't trust it.
Trust is a design problem, not an engineering problem. So is workflow fit. So is when to surface a recommendation and when to stay quiet. So is what the AI shows the user when it is uncertain.
Design thinking forces the team to ask the questions engineering, on its own, will skip:
- Who is the user? What is their job-to-be-done? What does their existing workflow look like minute by minute?
- What does success look like for them — in their words, on their day?
- What is the smallest version of this we can put in front of them in two weeks?
- What did they actually do, when we did?
- What did they say, versus what did we observe?
These are not luxuries. They are the questions that decide whether the model gets used or ignored.
Solo FDE vs. forward deployed AI team
What the lone-FDE model tries to compress into one person, the team distributes across four. The difference shows up in everything — pace, quality, what survives the next infrastructure migration, whether the customer renews.
What one person tries to cover
- Designs the problem framing while writing the code that solves it
- Owns architecture in the morning, integrations in the afternoon
- Drafts the success metric, then has no time to measure it
- Mediocre on three of four. Burns out. Customer routes around them.
What four roles cover together
- Designer keeps the team honest about the real problem
- Architect makes the system survive the next five years
- Engineers ship code customers' own teams want to keep
- FD product owns adoption — the only metric that matters
Where these teams are needed today
The team model concentrates in industries where AI delivers value through deep integration. The biggest hiring fronts in 2025 — click any industry for examples and comp range:
How the work actually splits
Inside the customer, an FD AI team's time tends to split roughly like this. Click any segment.
The team that includes a designer, an architect, two engineers, and a forward deployed product person ships outcomes a solo FDE cannot.
What the industry is still getting wrong
Six years after I put together that first team, the industry is mostly still hiring solo forward deployed engineers and hoping for the best. Five patterns I see repeat:
01Hiring one "FDE" and expecting them to be a team
One person cannot be designer, architect, engineer, and product manager simultaneously. The output will be mediocre on at least three of the four — almost always the three the engineer is least trained on.
02Skipping the designer because "design is fluff"
This is the single biggest mistake. Design thinking is the most predictive factor in deployment success I have measured. Without it, the team builds beautifully accurate solutions to the wrong problem.
03Skipping the architect because "we already have the model"
A model is not a system. A system is what the customer has to operate for the next five years. Without an architect, the pilot works and the production rollout collapses.
04Replacing the product person with a sales engineer
Same role title, different incentives. The SE optimizes for the next deal. The FD product person optimizes for this deployment's adoption. The two are not the same job and the substitution costs the customer the outcome.
05Treating the team as temporary
The teams I have seen succeed stayed embedded at the customer for nine to eighteen months. The teams that swap out after the first three months systematically underperform. AI deployment is not a sprint.
What this means for training
A role that did not exist five years ago has become a team that does not exist at most companies — and that no curriculum is producing.
The implication for the AI Impact Foundation is direct. We are not building a single "Forward Deployed Engineer" track. We are building four parallel tracks that produce people who can work together as a forward deployed AI team:
- Design for AI deployments — design thinking adapted for AI workflows
- AI systems architecture — the architect track
- Applied AI engineering — production-grade engineers
- Forward deployed AI product — the product side of FDE
A graduate of any one of these tracks can join a forward deployed AI team. A team that includes graduates of all four is the unit the market actually needs.
The lesson, six years on
The lesson from that first quarter has not gotten less true. The lone forward deployed engineer is a great hire and an incomplete answer. The forward deployed AI team is what enterprises actually need — and we are building the pipeline for it.
That's the work. Come help us do it.