Vol. IV — The operator
Michigan · USA
A medic's bias for outcomes, applied to AI.
Fourteen years in uniform, eleven of them in Special Operations. Then a master's in Artificial Intelligence and a second career building systems people depend on.
14
years in uniform
11
of them in Special Operations
18D
Special Forces medic
2025
M.S. Artificial Intelligence
The long version
Two careers with the same job description.
I spent fourteen years in the U.S. Army, eleven of them in Special Operations as an 18D Special Forces medic. The job is not what films suggest. Most of it is triage under time pressure with incomplete information, in conditions where the expensive failure mode is not indecision — it is confident, articulate wrongness that nobody questions because it was said with authority.
I earned an M.S. in Artificial Intelligence at the University of Michigan-Dearborn in 2025, and built across applied machine learning: computer vision, natural language, reinforcement learning, geospatial systems, and the considerably less photogenic work of getting a model to run reliably in production.
What I found is that AI projects almost never fail at the modeling. They fail because nobody senior is accountable for the outcome. Use cases get prioritized by who asked loudest. Evaluation is invented after the demo. Governance is a document written the week before an audit. The pilot stalls, and the postmortem blames the technology because the technology cannot argue back.
ClelandCo exists for the organizations that want that seat filled — someone who will own the roadmap, make the vendor calls, say which pilots to kill, and still be technical enough to implement the parts where a handoff would kill the work.
The other half of the practice is measurement. Buyers increasingly ask an assistant instead of opening a results page, and almost no business can tell you whether the assistant names them. That is a measurable thing, so I measure it — weekly, across four engines, with sample sizes and error bars attached, because a number without an interval is an opinion wearing a number's clothes.
I would rather tell you that nothing moved than tell you a story about why it did. That is the whole pitch, and it is the part of the medic job that transferred most cleanly.
How the work runs
Four steps, every engagement.
01
Diagnose
Find the gap using evidence, not vague AI or search claims. For AI visibility, that means measured surface and citation rates with sample sizes attached, plus the profile, schema, and directory data behind them. For advisory and fractional work, that means use cases, data, evaluation, governance, deployment, monitoring, and cost.02
Define done
Agree on the measurable outcome before anyone starts building. The most expensive projects fail when nobody defines success. Every engagement starts with a short, written definition of done, and a note on what would count as evidence that it happened.03
Build or fix
Do the smallest serious implementation that closes the gap. This keeps the work scoped, fast, and useful. No speculative platform rebuilds. No generic SEO package. Build the piece that moves the outcome and leave the rest alone.04
Handoff and monitor
Leave the system understandable, measurable, and operable. You get documentation, runbooks, next steps, and the instrumentation needed to know whether the system is working, including when the honest answer is that it is not.
Questions
Asked and answered.
- What is your background, exactly?
- Fourteen years in the U.S. Army, eleven of them as an 18D Special Forces medic, U.S. Army Special Operations, then an M.S. in Artificial Intelligence from the University of Michigan-Dearborn, completed in 2025. Applied work since across computer vision, natural language, reinforcement learning, geospatial systems, and getting models to run reliably in production.
- What does a Special Forces medic have to do with AI?
- The transferable part is not the setting. It is making decisions with incomplete information under a deadline, in conditions where the expensive failure is confident, articulate wrongness that nobody questions. That is also the failure mode of an AI function: a pilot nobody defined success for, an evaluation invented after the demo, and a postmortem that blames the model.
- Do you work with clients outside Michigan?
- Yes. The practice is remote and works across the United States. Michigan is where it is based, not the limit of where it operates, and every engagement is run over video, shared documents, and whatever your team already uses.
- Do you take employees, or is this consulting?
- Neither, quite. The fractional seat is a standing executive role held part-time — one to two days a week with decision rights and a budget line — and the intended ending is that you hire a full-time Chief AI Officer and I help you interview them. Consulting delivers a document and leaves; this is accountable for the outcome and plans its own exit.
Next
Enough about me. What's stuck?
Tell me what exists now, what needs to be true, and what is blocking the path. A short, specific note is plenty.