Writing
Essays on the systems, frameworks, and human dynamics that make — or break — product delivery at scale. (scroll for more)
Get blog alerts in your inbox.
Free — no paid account required.
My new book, Ace of Something: The Advantage of Being a Master of None, is out in paperback and on Kindle. It is a career memoir and a field guide for the person who was never the best at any one thing and ended up at the center of the work anyway.
Read essay →
For twenty years the PRD was a document humans read and interpreted. Now an AI agent reads it and ships. This post shows delivery leaders why product requirements must become an executable spec, and what changes when ambiguity turns into production risk.
How InvestmentThing works: a Claude strategist that proposes, a deterministic risk engine that overrules it, two brokers pretending to be one account, a second model that tunes the dials, and every expensive mistake I made getting there.
A practical walkthrough of building a systems atlas: the seven questions it has to answer, evidence based discovery instead of surveys, the schema that is the actual product, ownership as a social problem, decision records with a revisit condition, an interface where the relationships are the navigation, and the freshness protocol that keeps the whole thing alive after you stop looking at it.
Agentic AI is automating the base of the engineering pyramid, the junior work that teams were built on. That does not shrink your team evenly, it inverts it, pushing demand toward senior reviewers and architects while entry level roles collapse. Here is how to staff for the inversion without starving your own pipeline.
The context-engineering episode of Valuestream. Reliable AI agents fail on context, not prompts: what the model can actually see when it acts. The context budget, the four failure modes, and how to run context as a delivery discipline with version control, evals, ownership, and observability.
AI made writing code nearly free, but retiring it never got cheaper. That gap compounds into deprecation debt, a fast growing form of technical debt: dead paths, stale APIs, and un-migrated systems that quietly cap delivery throughput. Here is how to measure and govern it before it stalls your team.
Prompt engineering was never the real bottleneck. In production, agentic AI systems fail on context: what the model can actually see when it acts. Here is why context engineering is a delivery discipline, not a research problem, and how to run it with ownership, evals, and observability.
