Method / operating doctrine
AI in production: it's a matter of method.
The right stack is People → Workflows → Agents → Infrastructure. The model is a component, not the foundation. These are the 8 principles I design and ship with, distilled in the field.
Model ≠ Problem
If the agent fails, the flaw is in the design, not the model. The 95% of GenAI projects with no ROI (MIT NANDA 2025) don't have a capability problem: they have a design problem.
Trust is a technical precondition
The treasure is what experts don't articulate: the sub-optimal practices they've normalized. The diagnostic question: what would change your life that we've written off as impossible?
Data is a durable asset
Every piece of data you bring in and model well is a brick that stays yours. Connecting fragmented data reduces hallucinations and compounds a proprietary edge. That's the company brain thesis.
"Ignorant" AI-native redesign
Rebuild the process from zero under one constraint: one person + these AI tools. Automating the human org chart produces faster bureaucracy, not better systems.
The expert's signature, not the token lottery
Predictable quality via evals (tests on real scenarios) and pass^k, not after-the-fact review. The expert's standard lives inside the system, not applied afterwards.
The harness is the asset, the model isn't
Own the tools, contracts, evals, observability. The model is swappable: when something better ships, you switch in an afternoon. Leaderboards compress real differences; your evals don't.
Start from a high-value deliverable
Pick a task that takes weeks by hand, with strong validation and real business impact. Ship it in days. Measure the quality gain, not just the speed.
Multiplier, not a margin discount
AI should unlock previously impossible work. Treating it as cost-cutting erodes margins; treating it as expansion opens growth.