Playbook
How I run an AI transformation.
Six principles I keep coming back to, whether I’m working with a 70-engineer team or a 650-person business unit.
- 01
Throughput, not tokens
AI adoption isn’t the goal. I measure cycle time, shipped outcomes and quality, then work backwards to the tools.
- 02
Skills are the new internal libraries
Tribal knowledge becomes reusable skills and agent workflows. Write it once, and every engineer and every agent inherits it.
- 03
Evals before vibes
If you can’t score it, you can’t improve it. Every AI feature and every agent workflow ships with an eval.
- 04
Everyone ships
Typing speed was never the bottleneck. With the right guardrails, PMs, designers and ops ship real code, and engineers get their leverage back.
- 05
Guilds beat mandates
Adoption spreads through champions, demos and wins people can copy, not through top-down tool mandates.
- 06
Buy the frontier, build the moat
Rent the models and tools that move fastest. Spend your own engineering on the context, data and workflows only you have.
The operating stack
Fluent from the model layer to the org chart: hands-on where it counts, strategic where it compounds.
- Agentic development
- Coding agents
- Subagents
- Agent skills
- MCP servers
- Headless / CI agents
- Spec-driven development
- Models
- Frontier model selection
- Routing & fallbacks
- Prompt & context engineering
- Cost / latency tradeoffs
- Evals
- Golden datasets
- LLM-as-judge rubrics
- Regression suites
- Online quality metrics
- Product engineering
- TypeScript
- React
- Next.js
- Node.js
- Postgres
- Go
- Leadership
- AI guilds & enablement
- Org design
- Hiring loops
- Vendor negotiation
- Exec communication