The FDE labor market has never been more legible. Demand for the role grew over 1,100 percent year over year, the median posted base salary is around $174,000, seventy percent of postings include equity, and most of the roles are at AI-native companies with between eleven and two hundred employees. The market is broad, but so is the filter: the median posting attracts hundreds of applicants, and the person reviewing yours spends about forty-five seconds.
The four paths in
From software engineering: you have the production instinct and the comfort with unfamiliar codebases, and you’re missing the customer-facing muscle. Take three customer meetings a month in any capacity, build one substantial AI workflow end-to-end on real data, and rehearse the “they asked for X, I built Y” story with real examples.
From consulting or solutions architecture: you have the political mapping and commercial fluency, which are the hardest parts to teach, and you’re missing recent production code. Ship something real with the modern tools, get one merged PR into an open-source eval framework or agent SDK, and lean into the vertical your consulting career gave you.
From product management: you have discovery cadence and cross-functional judgment, and you’re missing hands-on depth in agentic systems. Do one AI build project every weekend for three months, and consider applying first to Agent Product Manager or Implementation Engineer roles; they’re calibrated for your background and tend to become FDE roles within a year.
From a failed startup: you are the most underappreciated candidate pool in the market. You have the founder-mode temperament and the customer-facing scar tissue that can’t be acquired anywhere else. What you need is an honest, metabolized failure story (not the press-release version) and one AI-native product shipped in the open under your own name.
The three-project portfolio
A resume makes claims. A portfolio gives a hiring manager something to check, and three projects is the right number, small enough to defend in depth and large enough to show range. One production AI workflow you’ve actually shipped, with real users and a measurable before and after. One written engagement postmortem, four to six pages covering discovery, the eval set, the first slice, the integration, and what you’d do differently. And one eval-set-plus-prompt-stack you can demo: a hundred labeled examples, a version-controlled prompt, and the wrapper that runs it on real data.
Package it in a public GitHub repo where the README survives the forty-five-second test: a plain-language first sentence in the customer’s terms, an honest impact paragraph, one architecture diagram, a runbook that actually runs, the documented eval set, and a two-minute demo video. If you don’t have a customer, build one. Find one person with one real problem, a friend’s small business, a nonprofit, a small SaaS founder, and run the engagement formally: scope it, build the eval set, ship the slice, run the loop for four weeks, and document everything. From a hiring manager’s perspective, the output is indistinguishable from a real FDE engagement.
The interview loop, and the stage that filters hardest
The standard loop has four stages: a recruiter screen, a technical screen (design an eval set, debug a production failure, or architect a deployment), a customer simulation, and an ownership conversation with the hiring manager. The customer simulation is the stage candidates rehearse least and lose most often.
It’s a role-play. The interviewer plays a senior customer stakeholder who insists on the wrong solution, or has to be told the validation results came back below threshold, or asks for a scope expansion that would break the timeline. The simulation tests composure under pressure when the right answer is hard to find, and the candidates who read as most credible are the ones who stay most relaxed, not the ones who push hardest. Take the customer’s stated solution seriously enough to ask what’s working about it. Name the underlying job rather than the surface request. Propose a path that respects the customer’s authority while letting the evidence decide; “let’s scope both versions and let the validation results choose” beats both capitulation and correction.
The final stage is a single question under the surface: do you have scars from shipping production systems in messy customer environments, and what did you learn? The good answer is granular, with a specific engagement, a specific failure, and the specific decisions you made in response. The hiring manager has had this conversation a hundred times and can tell a metabolized failure from a rehearsed one.
Ask your own questions too. The single most diagnostic one: do your FDEs ship code into your own product? The answer tells you whether the function is real or a cargo cult, and whether your career there will compound into product judgment or stall in services-engineer mode.
Want the full playbook, with the templates, the interview transcripts, and the week-one checklist? It’s all in my book Forward Deployed AI Engineering: A Working Guide to the Hottest Job in Software, available on Amazon: [Amazon link]