The 90-Day FDE Path.
From software engineer or data scientist to Forward Deployed Engineer — the role OpenAI, Anthropic, and all other AI companies are betting billions on. This is the same path I coach from.
Kris Diallo — Forward Deployed Engineer. Written from the field. Strip away the hype — forward deployment is three skills.
The Forward Deployed Engineer walks into a company, scopes the problem, and ships the AI that solves it. Palantir made the role famous; OpenAI, Anthropic, Google, and Databricks are hiring for it now. Underneath the title:
Understand the business.
Walk into an industry you’ve never worked in and figure out how it makes money. Read an org chart and know who owns the problem, who feels the pain, and who signs the check. Speak in the metrics they care about — churn, margin, cycle time.
Turn understanding into architecture.
Take a messy human problem in, and output an optimized system design that improves a number the business cares about. The skill is in the translation.
Build and deploy.
One continuous cycle: build → test → deploy → monitor. Capture every fix as an eval, ship only tested changes, and turn every failure into a better version.
If you’re a working SWE or DS, you already have skill 3. The gap is skills 1 and 2 — and that gap is why the role pays $200–400k+ while pure coding roles get automated away. Skills 1 and 2 are pure human judgment. Automation eats assignments, not alignments.
Most engineers are not as good with AI as they think.
Before the business skills matter, there’s a more important gap to cross. Here’s the AI skill ladder — be honest about where you sit:
Most people sit between rungs 1 and 2. The 90 days below move you up the ladder while you build proof.
The 90 days.
Master the machine, find a real problem.
Work context engineering, tool calling, and agentic loops until they’re reflexes — build small things that work, every day. Then start skill 1: find one place where real people feel real pain, and interview them. Discovery is detective work: a process lives in someone’s head as an undocumented decision tree, and your job is to extract it.
Design it, build it, put it in someone’s hands.
Write the architecture doc around the pain point, rooted in the metric it should move — nobody cares about the stack. Build v1 with whatever ships fastest. The only goal: a real person using it for real work, with the before and after measured.
Production-harden, write it up, land the seat.
Evals, fallback paths, access control, a rollout plan that would pass a security review. Write two case studies the way an FDE reports to a client: problem → design → what broke → measured outcome. Then reposition everything — resume and LinkedIn lead with outcomes — and run both plays: apply to AI-native companies, and pitch your company on making you the person who deploys AI internally.
Want help running it?
The path is free and always will be. If you want to run it with me, 1:1 — here’s how coaching works, or book the call directly and we’ll map your 90 days together.