Start with the decision
A feature matters only if it helps someone make a better decision, complete a workflow, or remove a measurable constraint.
Product manager · Independent builder
Good product work starts before the solution.
I work through messy workflows, competing constraints, and uncertain evidence to find what is worth building—then determine where AI creates real leverage and where it does not.
A two-minute product decision
Don’t ask, “Where can we add AI?”
Ask what needs to change in the workflow. Pick the first move you would make.
A team loses hours every week moving information between disconnected tools.
Leadership wants an “AI solution.” The process is inconsistent, exceptions are common, and nobody agrees on the baseline cost.
What should happen first?
How I would reason about it
My first move
Find the repeated decisions, exceptions, data gaps, and cost of the current workflow. That tells us whether the right answer is AI, ordinary automation, process redesign—or nothing at all.
Then I’d testFrequency × effort × consequence
How I think
Product judgment, made visible.
I’m less interested in displaying a toolkit than showing the choices the toolkit helps me make.
Make tradeoffs visible
Prioritization is not a score hidden in a spreadsheet. It is a shared explanation of what matters now, what waits, and why.
Design for the failure
Especially with AI, the product is not just the happy path. Confidence, review, traceability, and recovery are part of the experience.
Independent product work
Two questions I’m actively building around.
Working products now. Deeper case studies next.
01
Where could AI create measurable value—and what must be true before investing?
An evidence-grounded assessment platform that turns workflow documentation into reviewable AI opportunities, with transparent scoring, explicit uncertainty, and human decisions preserved.
- Evidence before recommendations
- Deterministic ranking over black-box scores
- Human evaluation and revision history
02
What would responsible AI decision support look like in a regulated workflow?
A 0→1 prototype for source-grounded compliance research and case analysis, designed around traceability, human review, and an audit trail rather than autonomous decisions.
- Source-grounded regulatory search
- Human-in-the-loop case review
- Auditable inputs, outputs, and decisions
This is version one
The work is live.
The case studies are taking shape.
I’m documenting the decisions, tradeoffs, and evidence behind these products as I build them.
Follow the work on GitHub ↗