Methodology
Four phases. No deck-only deliverables.
Every engagement runs the same loop. Strategy and engineering are not separate tracks — they happen at the same table, with the same person accountable end-to-end.
Diagnose
Understand the business before touching the technology.
Two-week immersion with leadership, ops, and engineering. I sit in operating reviews, talk to the people doing the actual work, and read the data.
Output is a problem statement the CFO can read — not a buzzword bingo card. Where is value leaking? What would ‘done’ look like in dollars?
If the answer is “you don’t need AI for this” — that’s the answer. I’ll say so.
Design
Architecture, change plan, and economics together.
Solution architecture, change management, and unit economics get designed at the same table by the same person. The big-firm pattern of separate strategy and engineering tracks is exactly how things break in production.
Build-vs-buy is decided per workstream, not as a religion. Vendor choices include an explicit escape hatch — you should never be one model deprecation away from a rebuild.
Out the door: an architecture you can defend to a board, and an integration plan your engineers can defend to themselves.
Deliver
Ship in 90-day increments. Every increment ends in production.
No twelve-month waterfalls. Every 90 days produces a system that’s actually serving real users or real workloads, with metrics on the wall.
Evaluation is non-negotiable. Before agents touch a customer, there’s a regression suite and a guardrail set. Before models touch a regulated workflow, there’s an auditable trail.
Production or pause: if the metrics aren’t moving by the next checkpoint, we stop and replan. No escalation game, no scope creep dressed as ‘phase two.’
Durable
The system survives my exit.
Every engagement ends with documentation, runbooks, evaluation harnesses, and a trained internal team that owns the system going forward.
Governance lives where it can be enforced — in the deployment pipeline and the procurement process — not as a PDF on a shared drive.
Six months after I leave, the system should be more useful, not less. If it isn’t, that’s a delivery failure, not a customer one.
Principles
What I won’t do. On purpose.
I won’t sell you AI for AI’s sake.
If a spreadsheet, a process change, or hiring two people would solve it — I’ll tell you. The fastest way to lose a long-term client is to ship them an LLM where they needed a checklist.
I won’t hand off to a junior team.
When you hire me, you get me. I scale by being selective about engagements, not by running a pyramid of associates.
I won’t lock you to one vendor.
Every architecture I design has an escape hatch. Model providers will keep shifting. Your business shouldn’t shift with them.
I won’t leave without runbooks.
If the system breaks at 2am six months after I’m gone, your team should know exactly what to do. That’s a delivery deliverable, not a documentation hobby.