How To Become a FDE

Came across this awesome post by Maryam on LinkedIn which puts together succintly exactly my view on how to become a forward deployed engineer.

🛠️🧭 How to Become a Forward Deployed Engineer for Production AI Agents.
Forward Deployed Engineering is not AI Engineering at a customer site.
𝗠𝘆 𝟳-𝗦𝘁𝗲𝗽 𝗿𝗼𝗮𝗱𝗺𝗮𝗽 𝗳𝗿𝗼𝗺 𝗰𝘂𝘀𝘁𝗼𝗺𝗲𝗿 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝘁𝗼 𝗺𝗲𝗮𝘀𝘂𝗿𝗮𝗯𝗹𝗲 𝗔𝗜 𝗶𝗺𝗽𝗮𝗰𝘁.

It sits between:
– customer problems
– product engineering
– production delivery
– business impact

It is closer to an acting CTO.
Enough technical depth and Enough business judgment to choose what is worth building.

Here is the roadmap I would follow:

》𝗦𝘁𝗲𝗽 𝟭: Build Full-Stack Depth
✸ Learn Linux, frontend, backend, APIs, databases, and authentication.
✸ Be able to trace a failure across the system.
✸ An agent is useless if it cannot fit into the customer workflow.
→ Example: Connect a claims agent to the policy database and review screen.

》𝗦𝘁𝗲𝗽 𝟮: Master DSA & System Design
✸ Understand algorithms, services, queues, state, and failure paths.
✸ Plan for scale before adding more agents.
✸ Do not hide weak architecture behind a strong model.
→ Example: Make tool calls idempotent before allowing automatic retries.

》𝗦𝘁𝗲𝗽 𝟯: Add AI Engineering
✸ Learn RAG, memory, tool calling, structured outputs, guardrails, and evals.
✸ Define success with users before delivery.
✸ Do not measure quality by whether one demo worked.
→ Example: Build a labeled eval set with domain experts.

》𝗦𝘁𝗲𝗽 𝟰: Learn Production Delivery
✸ Use cloud, infrastructure as code, CI/CD, security, and observability.
✸ Trace what the agent saw, decided, called, and changed.
✸ Add rollback and human approval for risky actions.
→ The customer must trust that the system will not break.

》𝗦𝘁𝗲𝗽 𝟱: Discover the Real Problem
✸ Sit with users and map the current workflow.
✸ Find the bottleneck with the most business value.
✸ Deliver the smallest end-to-end unit that proves impact.
→ Do not start with “Where can we add an agent?” Start with “What is blocking the customer?”

》𝗦𝘁𝗲𝗽 𝟲: Measure Business Impact
✸ Connect the build to time saved, errors reduced, revenue, or delivery speed.
✸ Balance scope, speed, and quality.
✸ Stop measuring success by token usage.
→ A busy agent is not the same as a useful agent.

》𝗦𝘁𝗲𝗽 𝟳: Close the Product Loop
✸ Write clear field notes and communicate tradeoffs.
✸ Bring repeated customer failures back to product teams.
✸ Turn one-off fixes into reusable platform capabilities.
→ The next deployment should be better because of the previous one.

𝗣𝗹𝗲𝗮𝘀𝗲 𝗥𝗲𝗺𝗲𝗺𝗯𝗲𝗿:
A Forward Deployed Engineer is not an AI consultant with slides.
And not only an AI engineer writing custom code.
They discover, build, deploy, measure, and turn one customer problem into a better product for many customers.