Career roadmap
Forward Deployed Engineer (FDE)
Embed with a customer, build working software inside their environment, then turn what you learn into product.
Before you start FDE
- Full-stack shipping ability in any stack
- SQL and API integration experience
- Docker and one cloud
- Willingness to speak directly to customers
Core engineering depth
Be the person who can build the whole thing alone if needed.
One language, deeply. Enough to be the only engineer in the room.
Ch 26 — REST Best Practices- API design and versioning
- Auth and sessions
- Background jobs
- Error handling and logging
You will design a schema on day three of an engagement.
- Normalisation trade-offs
- Indexes
- Migrations under load
- Reading someone else's schema
Their environment will not match yours. Ship something portable.
- Containerising an app
- CI pipelines
- Config and secrets per environment
- Logs you can read from outside
The most under-taught FDE skill. You will be dropped into strange repos constantly.
- Entry points and request paths
- Finding the data model first
- Debugging without a local setup
BuildA small end-to-end app deployed for someone other than yourself, with auth and a real database.
Integration craft
Customer data is never clean and their network is never open. Work anyway.
Excel serial dates, lakh-formatted numbers, trailing spaces, three date formats in one column.
- Tolerant parsers
- Per-row rejection reports
- Encoding problems
- Never silently dropping a row
You will re-run the same import three times in one afternoon.
- Upsert on a natural key
- Safe reruns and partial failure
- Backoff and rate-limit handling
Their identity team decides whether your app ships.
Ch 44 — Authentication- SAML and OIDC basics
- SCIM provisioning
- Role mapping
- Debugging a broken assertion
No outbound internet is a normal constraint, not an edge case.
- Private networking
- Offline installs and mirrors
- Data residency
- Working with their security review
BuildAn importer that ingests a deliberately dirty 50k-row spreadsheet and reports exactly what it rejected and why.
AI application layer
Most FDE work today is putting a model against a customer's own documents and processes.
Enough of the AI Engineer track to ship a real feature.
Ch 89 — AI Engineer (stages 1-3)- Structured outputs
- Streaming
- Cost per request the customer will accept
Their contracts, runbooks, and policies — with citations, because trust is the deliverable.
Ch 80 — RAG Pipeline- Ingesting their formats
- Permissions-aware retrieval
- Citations
- Refusing when unsure
An eval set the customer wrote is also the acceptance criteria for the pilot.
- Collecting real cases in discovery
- Getting sign-off on the rubric
- Reporting results to non-engineers
BuildA retrieval assistant over a real organisation's documents, with an eval set that organisation agrees with.
Customer skills
The real differentiator. Most engineers never build this, and it caps their level.
Find the workflow with real hours or money attached. Watch someone do it.
- Questions that surface cost
- Shadowing a user
- Separating stated from actual problems
A demo in two weeks beats a plan for two quarters.
- Cutting scope without cutting value
- One agreed success metric
- Writing the scope down
Show it early, ugly, and working. Polish signals distance from the problem.
- Framing a rough demo
- Handling live failure gracefully
- Capturing feedback in the room
Every yes is a commitment someone will hold you to on Friday.
- Trade-off framing
- Offering the near thing instead
- Escalation without blame
A pilot without a metric never ends and never converts.
- Success criteria up front
- Weekly check-ins with evidence
- Deciding to stop
BuildPick a real small business, find one painful workflow, ship a tool for it, and measure hours saved per week.
From bespoke to product
The job is not consulting. What you learn in the field becomes the product.
Three customers asking differently for the same thing is a feature, not three projects.
- Comparing engagements
- Separating config from code
- Saying which 80% generalises
You will leave. What you wrote down decides whether the deployment survives.
- Runbooks for their ops team
- Support escalation paths
- Known limitations documented
Field evidence beats opinion in a roadmap meeting — if you bring numbers.
- Writing a product ask with evidence
- Quantifying customer pain
- Prioritising against other customers
BuildA written case study: the customer problem, what you shipped, the metric it moved, what should become product.
FDE tools on your CV
- TypeScript or Python
- Postgres and one warehouse
- Docker, Terraform basics
- LLM APIs
- Postman / curl and devtools
- A notebook for customer notes
What FDE employers ask to see
- A shipped tool a real organisation uses weekly
- A case study with a before/after metric
- Evidence you handled bad data and a locked-down environment
The fastest-growing hybrid role in AI companies and the hardest to fill. Half product engineer, half consultant — but you write the code and own the deploy. Also advertised as solutions, deployment, implementation, or customer engineer.
Content last reviewed 2026-08-31. Guidance only — no institute or paid placement is endorsed anywhere in this book.