Leadership

How I run an AI organization

Six operating principles, each paired with where I have applied it. The case studies carry the full detail.

  1. 01

    Strategy carried through to delivery

    I own AI work from problem definition through production deployment. A strategy that stops at a roadmap leaves the hardest part to someone else: evaluation, integration, and the people who have to operate the result.

    In practice

    Procured three consecutive DoD contracts and carried the defect-recognition system through to DoD approval for shipyard integration.

    See the evidence for Strategy carried through to delivery

  2. 02

    Innovation with safeguards built in

    Research capacity is protected, and every system answers to an evaluation harness and governance controls defined before it ships. The harness is what lets a team change a model quickly without guessing what the change cost.

    In practice

    Kept about a third of ORBIS AI capacity on research; established model-access, environment-segregation, and human-oversight practice across the DoD portfolio.

    See the evidence for Innovation with safeguards built in

  3. 03

    Strong teams with real ownership

    I practice servant leadership: I own priorities, resources, customer access, and organizational friction, and engineers own their areas of the work. I hire for depth plus range so people can follow a problem across data, model, integration, and domain validation.

    In practice

    Grew the ORBIS AI department from one engineer to a core team of ten and led programs of 30+ across departments, contract experts, and government partners, directing the project managers who ran them; rated 5/5 for five consecutive years.

    See the evidence for Strong teams with real ownership

  4. 04

    Executive trust earned with evidence

    Headcount and budget follow demonstrated value. I document the path from research experiments to funded programs so leadership can see which earlier investments produce current revenue.

    In practice

    Worked directly with the ORBIS C-suite on AI funding and headcount; copyrighted AI/ML IP from protected research became the main source of ORBIS R&D revenue.

    See the evidence for Executive trust earned with evidence

  5. 05

    Adoption designed in from the start

    The people operating an AI system can refuse it by routing around it. Which findings surface, how confidence is shown, and where a human makes the call get as much design attention as the model.

    In practice

    Designed the radiographic inspectors' hand-off for the DoD defect-recognition system, which changed what they spent their attention on while leaving the decision with them.

    See the evidence for Adoption designed in from the start

  6. 06

    Leadership distributed before it is needed

    Shared context stops scaling as a team grows. Standards, decision rights, and institutional knowledge are written down early, and additional leaders are developed deliberately so the team can carry work forward without me in every decision.

    In practice

    Documented evaluation practices, deployment requirements, security controls, and architecture decisions as the ORBIS team approached ten engineers.

    See the evidence for Leadership distributed before it is needed

Decision framework

How I assess an enterprise AI investment

Four questions decide most AI funding calls before any model is chosen: whether the value is real, whether the work is feasible inside your constraints, what a wrong answer costs, and whether the people who must operate it will.

Illustrative only. This is how I frame an AI investment decision; it is not drawn from a client engagement.

Value
  • Which decision or workflow changes, and what does a measurable result look like?
  • Does it remove a bottleneck, improve a decision, or do work at a cost, speed, or scale the current process cannot match?
Feasibility
  • Is the data available, and is it allowed to move where the model runs?
  • Is the task bounded, with checkable outputs, or does it genuinely need frontier-scale capability?
Risk
  • What does a wrong answer cost, and who catches it before it matters?
  • Is the evaluation standard agreed before development, and is the security boundary designed first?
Adoption
  • Who operates the result, and where does review fit into a day that is already full?
  • Who owns the system, and its evaluation harness, after handoff?

Rate all four dimensions to see how I would frame the funding decision.