FDE vs SWE, SE and AI Engineer

Three job titles sit close enough to FDE on a careers page that reading them carelessly will cost you time, money and a well-prepared loop against the wrong rubric. This lesson is about telling them apart.

The thing to carry through all of it: the test is who owns what, after whom, and with what access to the customer's codebase. Every other difference falls out of that one.

The four roles, laid out

The point of looking at all four at once is that the overlaps are where candidates get lost. An FDE and a Solutions Engineer both travel and both talk to executives. An FDE and an AI Engineer both write LLM-shaped code. An FDE and an SWE both ship software into a repository. Lean on the differentiators, not the overlaps.

The one test that cuts cleanly

When a job description is ambiguous, apply this test in order:

1. Step 1

   When does the role start?

   Before the contract is signed: Solutions Engineer. After the contract is signed: one of the other three.

2. Step 2

   Whose repository does the code land in?

   Customer's repository: FDE. Employer's repository: SWE or AI Engineer.

3. Step 3

   Is the work horizontal or bespoke?

   Every customer sees it: SWE or AI Engineer. One customer sees it: FDE.

4. Step 4

   Is LLM craft the spine of the job?

   Yes, and the job is bespoke: FDE. Yes, and the job is horizontal: AI Engineer. Not especially: SWE.

The failure mode of this test is a role that actually belongs to two categories. That is more common than you would hope. An "Applied Forward Deployed AI Engineer" is somebody's attempt to say all four at once, and the test does still work — ask when the work starts, and whose repo it ends up in, and the answer will pick one. The title is marketing; the work is the test.

What you would and would not do in each

A concrete walkthrough of the same working day, split four ways. Say the customer is a mid-size bank, and the project is an agent that helps their compliance team flag suspicious transactions.

The Solutions Engineer builds a demo on sample data, co-presents it to the bank's CTO, writes the technical section of the RFP response, negotiates the proof-of-value scope with procurement, and hands the signed contract to delivery. Their fingerprints are on the deal, not on the production code.

The FDE takes the handoff. They sit with the compliance analysts for two days, agree what "suspicious enough to flag" measures against, build a retrieval layer over the bank's transaction history, define the eval rubric, ship a pilot to a four-analyst team, and get paged when the Monday morning false-positive rate spikes because a new merchant category entered the data. Their fingerprints are on the production system.

The SWE at the same employer — say, on the agents infrastructure team — owns the control-flow primitives that let the FDE's agent make a tool call reliably. The compliance use case is one of dozens their code is running behind. They have never met the bank.

The AI Engineer at the same employer — say, on the evals product team — owns the evaluation harness that the FDE is wiring into the bank's CI. They also serve every other customer. They get a shared Slack channel with FDEs across engagements because the FDEs are their most demanding users.

All four jobs can be satisfying. Not all four jobs are the same career. Reading a JD against this split is what tells you which one you are interviewing for.

When the titles blur in practice

Three edge cases worth naming, because the market is not tidy.

"Deployed engineer" at a smaller company. Often an FDE in everything but the specific noun. Apply the test above; if the work starts after the contract and the code lands in the customer's repo, treat it as an FDE role for study purposes.

"Customer engineer" at a hyperscaler or cloud provider. Historically a hybrid of Solutions Engineer and FDE. Read the JD for the on-call line and the production-ownership line; if both are present, it is FDE-shaped work.

"Field CTO" or "Field Principal Engineer". These are FDE roles at the senior end, usually with an architect posture and less day-to-day code. The LLM craft is still expected, but the thing being evaluated is judgement at scale.

What to take away

Four roles — FDE, SWE, Solutions Engineer, AI Engineer — all adjacent and regularly confused. The clean cut runs through two questions: when the work starts relative to the contract, and whose repository the code lands in. The FDE is the only role that starts after the contract and ships into the customer's repo, which is why the role carries the on-call and the LLM craft and the production ownership all at once. Reading JDs with this split in hand is how you spend a week preparing for the right loop.

Next, Part 2 begins, and the subject changes from the shape of the role to the material it is made of: tokens, context, cost and the honest limits of sampling.