Written for leaders who must set rules and fund work: how capability, risk, and organizational reality interact, and what that implies for standards, oversight, and adoption.
Deliberate refactoring as an enabler for autonomous refactoring. Splits continuous refactoring under AI into two workstreams: deliberate (human-led structural work that makes a codebase AI-friendly) and autonomous (agent-led work that targets structural drift). Autonomous refactoring runs as two practices, systematic refactoring that enforces the codified standard and opportunistic refactoring…
Why nanoservices, once an antipattern, are the architecture AI-driven development requires. Re-examines Arnon Rotem-Gal-Oz's 2012 nanoservice antipattern definition (SOA Patterns) and shows it was a verdict on economics, not granularity. Walks through the four original objections — poor performance, fragmented logic, development overhead, management overhead — and argues three have…
A theoretical grounding based on Anderson (1972) and Prigogine (1984). Applies Anderson's 'More Is Different' to the IT stack and Prigogine's bifurcation framework and dissipative structures to the architectural transitions from mainframe to AI-driven, built from original sources as the foundation for the AI nanoservices argument.
If you no longer read what AI builds, you have to start building smaller. Argues that non-deterministic builders need units small enough to verify autonomously, and that the overhead which made nanoservices an antipattern has collapsed. Introduces permanent absence, verification replacing review, and the engineer's role shifting to specification, architecture, and templates.
Entry-level tasks disappear, but entry-level careers are wide open. Applies Shell Theory to the junior developer path: AI takes the well-defined tasks at the flatline, but the career runs through evaluation and judgment, not boilerplate. High-agency juniors use AI as scaffolding to reach the amplification zone; low-agency juniors become shells dependent on output they cannot evaluate.
How to retain your most capable engineers while AI intensifies work. Maps how AI work intensification — technostress, Jevons paradox, addictive feedback loops — creates a retention risk concentrated in the engineers you can least afford to lose, and outlines intensification guardrails, targeted investment in high-agency juniors, and structured autonomy as the response.
A framework for which context engineering practices remain as LLMs improve. Distinguishes intentional context engineering (durable workflow expression) from compensatory context engineering (temporary workarounds for current LLM limitations), applied across Cursor rules, commands, skills, hooks, and subagents.
A critique on viral essays written by AI. Examines the ‘Something Big Is Happening’ thesis using SWE-bench data to argue for an S-curve interpretation rather than unbounded exponential growth, and offers Shell Theory as a more moderate alternative.
Using the human brain to predict how LLMs will evolve. Maps current LLM limitations (attention decay, lack of persistent memory, brittle reasoning) to the context engineering practices that compensate for each, and identifies which scale advances will obsolete those practices.
A model that explains when AI replaces programmers. Resolves the apparent contradiction between AI raising every developer’s baseline, amplifying skilled engineers, and eroding critical thinking, via the AI Flatline, the amplification zone, and the agency feedback loop.