I found out I was a Forward Deployed Engineer by building the system that defines one.
A large enterprise technology company set out to stand up an FDE function for a role the market could not agree on. I built the multi-agent system that mapped the whole market and defined the role, then read what it produced and recognized myself in it. The method was the answer.
That is the job. Go inside the problem, understand it on the customer's terms, and build and ship the thing that solves it. Every system I build runs the same method:
And into every build, a layer that never invents a person, a signal, or a number. For a decision that matters, an unsourced answer is worthless.
I am not a traditional software engineer, and I do not pretend to be. What I bring is the combination the role is built on. I have shipped agents into production, and I have spent years on the post-sales side of enterprise implementation, including global rollouts with Big Four partners. Most people have done one. The value is doing both: the most technically credible person in the business review, and the most commercially grounded person in the design session.
A large TA org cannot see its own posting health: how many are live, which are broken or non-compliant, whether search and AI engines can even read them.
An audit is mostly facts, not opinions. Split the work by fact versus judgment: a deterministic engine proves what it can, an interpretive layer handles the rest.
Standard-library Python. Detects the ATS, maps every posting, runs deterministic checks, and clusters findings to the upstream config cause. Every finding cites a req and a URL.
Run free against thousands of live reqs across six ATS platforms. On a Fortune 500 medical-technology company's ~1,130 reqs it surfaced real, verifiable errors and traced repeated ones to single fixes.
The client needs an early signal for distressed airframes, from public data, with no scraping stack and no guesswork.
Reverse-engineer the signal from FAA notices already published when an aircraft disables a runway. Deterministic wording, low false positives.
A monitor keyed on the notice id: airport, runway, real duration from the cancellation, and a one-page incident brief when a closure crosses a threshold.
A verified go, ~50 to 150 qualifying events a year. I caught and cut a convenient source that turned out not to exist before it became load-bearing.
The roles worth wanting increasingly have no agreed title, and the search becomes a defeating, title-driven grind. You cannot search a board for a title the market has not named.
Run the search like a build. Make the seeker's version of good explicit, ground it, and reverse-engineer the path, the same method I run for enterprise customers, pointed at the candidate.
A single self-contained page. One starter prompt hands any AI the whole system: a grounding doc and career blueprint, a daily agent that scores roles by fit against an anchored rubric, one-size-fits-one resumes, an application pack, and a weekly loop that sharpens as you go.
A working system a stranger runs for free, and my own daily search, now running two shapes of good at once. Built in public and adapted freely.
More builds are on the way.
In 2020 a head injury erased my memory. I rebuilt my house and my brain in the same place, with my hands, on a street called Whiterock Road. I came to AI without the sci-fi story. I saw a tool and started with the friction. I do not carry the inherited library of how a problem is supposed to be solved, so in a space the market has not finished defining, I am not guessing at a playbook. I am building the right one. The longer version is at whiterockroad.org.