Writing
Practitioner essays on building and architecting systems: architecture, delivery, AI, and transformation. Where the discipline holds, where it breaks, and what it costs.
October 18, 2026
Buffers and shapes
Give our data rows and columns, make ownership explicit, and add matrices on the GPU.
Read essayEarlier essays
21 essays15 min read
Let's Build a Transformer! · Part 1
The cradle
Find a GPU, run our first C# kernel, and check every result before building on it.
7 min read
Let's Build a Transformer! · Part 0Prologue: Getting the cradle ready
Set up the tools, build the starting program, and check the dependencies we’ll use in the first lesson.
13 min read
Transformation Under Scale · Part 5Transformation Under Scale — Part V: Derive the Roadmap
A transformation roadmap is not drawn from preference. It is the operating model laid out in time: the dependencies between capabilities are facts about how the business works, the …
6 min readThe Case for Craft
Moving from writing the code to directing the people who write it does not cost you the craft. The craft was never the keyboard. It is the model you reason with, and it only decays …
9 min read
Transformation Under Scale · Part 4Transformation Under Scale — Part IV: Write Authority
A capability's boundary is the set of facts it alone writes. Draw the line by write authority, then move the pen from the legacy system to the target one, one boundary at a time, …
10 min readReasoning Tokens Steer the Trajectory
The reasoning-token debate is pitched at the wrong level. A reasoning token steers a trajectory through the model's probability field; what it means is incidental to what it does, …
10 min read
Transformation Under Scale · Part 3Transformation Under Scale — Part III: Read the Estate
You cannot bound what you have not read. Before any target state, read the estate at the right depth: rank the evidence, label the inferences, and let the reading size the plan.
6 min read
Transformation Under Scale · Part 2Transformation Under Scale — Part II: Transform the Company, Not the Technology
A transformation re-draws the company's operating model, who owns which business capability, not its technology. The architecture is the downstream expression of the ownership …
6 min read
Transformation Under Scale · Part 1Transformation Under Scale — Part I: Constrain the Possible, Not the Ideal
A transformation does not begin by designing the ideal system. It begins by turning the company's objectives into constraints and holding them over the estate you already have.
10 min read
Governance Under Scale · Part 4Governance Under Scale — Part IV: Model Selection, NAIC, and the Crosswalk
The NAIC AI bulletin, NIST AI RMF, ISO/IEC 42001, the EU AI Act, and SR 11-7 all require a compliant system, not a compliant model. The crosswalk shows they are one control …
7 min read
Governance Under Scale · Part 3Governance Under Scale — Part III: Revocation and the Reachable Decision Surface
If governance requires the ability to constrain authority, then the critical question is not how a system behaves, but what it remains permitted to do.
7 min readWhen AI Systems “Dream”: A Failure of Architecture, Not Models
Many AI failures labeled as hallucinations are actually coherent systems operating without sufficient grounding or constraint. The fix is architectural, not model-based.
5 min read
Governance Under Scale · Part 2Governance Under Scale — Part II: Monitoring Is Not Control
Monitoring provides visibility into system behavior. Governance requires the ability to constrain it.
4 min readAgents Are Actors With Intent, Not Guarantees
AI agents are not a new category of system. They are actors with inference where the decision used to be, which reintroduces familiar distributed systems problems under …
5 min readGhost in the Machine: Adversarial Priors in AI Systems
Large language models learn the statistical structure of the text they are trained on. Because written language overrepresents conflict, persuasion, and strategic reasoning, the …
7 min read
Governance Under Scale · Part 1Governance Under Scale — Part I: Human Override Is Not Governance
Human-in-the-loop is commonly treated as a safety guarantee. Under scale, it becomes a delegation surface, and one of the primary vectors of governance drift.
7 min readAI in Regulated Systems: Where Architecture Becomes Governance
In regulated systems, the primary AI risk is not hallucination. It is allowing probabilistic inference to directly mutate deterministic state.
4 min readAI Is Increasing Your Delivery Velocity While Moving Your Problems Downstream
AI speeds up delivery. In many teams, it also delays the moment where real structural problems surface.
7 min readKnowability
How systems preserve the connection between the rules they enforce and the outcomes they produce as they change and interact.
4 min readWhen Inference Becomes Control
The late-2025 fix for an unreliable agent was more agents. But you cannot monitor your way to control over a system that runs on inference, because inference discards the evidence …
3 min readAI Is a Delivery Tool, Not a Strategy
AI accelerates delivery when governed; without standards it creates fragility.