The New Rules
The methodology on this site rests on a single body of thought: The New Rules, a survival manual for developers in the age of AI. It was written by practitioners, not consultants. Its question is the one every engineering organization now faces: when machines write code, what do humans do? Its answer is a working discipline built on the shift the book names - from the old rules of mastered frameworks, stacked features, and accumulated years to the new ones: architect systems, speed to value, judgment and taste, simple solutions, shipped products.
This page is the manual's working counterpart. The writing collected below applies that same discipline to the builds leadership actually funds. It spans assessment, strategy, validation, and measurement for the questions a regulated, capital-intensive organization faces when it commits to AI. The book itself lives at its own address; read it there.
Recent writing
The pieces below apply the same operating discipline to the builds leadership actually funds across assessment, validation, and the metrics that survive go-live. Each piece specifies the vertical it was written for and the delivery stage it grounds, connecting the discipline directly to engineering practice.
-
The capacity and control a private AI stack delivers in operation
Once a private AI stack is live, its gains reduce to two properties an evaluator can inspect: capacity as owned compute bounded by measured floors and ceilings and set against metered token prices, and control as governance over data, models, configuration, and change. Neither depends on a provider's calendar. Both show up in operation, on a cadence, in evidence the organization itself generates.
- Technical Publication
- Technology
- Finance
- Building and Validation
-
What organizations gain when AI runs inside their network
Most organizations are told that the value of AI and the custody of their data are a trade. When inference runs inside your network on hardware you own, the trade disappears: whole categories of risk and administrative work stop existing, and gains in capacity, auditability, and independence start compounding in their place.
- Blog
- Finance
- Legal
- Strategy and Planning
-
Sizing an on-premise GPU configuration against the real workload
A defensible on-premise GPU configuration is the intersection of three measured budgets: latency, throughput, and key-value-cache memory. It is determined by replaying an organization's actual workload, not by relying on a vendor reference architecture.
- Technical Publication
- Technology
- Finance
- Building and Validation
-
The metrics that make an AI investment defensible after go-live
Go-live marks the beginning of operational accountability. Verifiable performance metrics reviewed by audit committees are what sustain long-term funding for enterprise AI investments.
- Blog
- Finance
- Measure and Improve
-
Validating AI on real hardware before the procurement commit
A model that performs well in an experimental notebook often fails in production. Validating performance on physical target hardware against real enterprise workloads provides the only defensible baseline for procurement.
- Blog
- Technology
- Finance
- Building and Validation
-
One month into the Hefty beta, a field report from the polish phase
A month ago I reported what I learned taking Hefty from zero to an open beta with AI as my primary coding partner. This is the field report from the month after, and the question has changed: not whether AI can help build the product, but whether one person can sustain it.
- Blog
- Technology
- Building and Validation
-
What I learned shipping the Hefty beta with an AI coding partner
The open beta of Hefty, a local-first AI agent that controls your computer, automates workflows, and manages files, is live. I built it from scratch with AI as my primary coding partner, and the build confirmed the argument I first made in The New Rules: when implementation speed approaches zero, thinking speed becomes everything.
- Blog
- Technology
- Building and Validation
-
The assessment that earns an AI roadmap its board mandate
Most AI programs begin with ungrounded ambition and stall in their first quarterly review. A rigorous readiness assessment separates surviving initiatives from failed pilots.
- Blog
- Finance
- Legal
- Assessment