Ideas, experiments, and the occasional useful rabbit hole.
A friendly, evidence-led guide to why successful AI-assisted feature work can leave repository residue and how to prune it with two independent model reviews, small reversible changes, and validation of the actual diff.
Read the article →US chip controls may reward Chinese AI ingenuity while still slowing progress. The evidence supports that tension; an AGI breakthrough remains an open possibility.
Read the article →Stronger models do not automatically make better products. The defaults around them need owners, evidence, and a reason to change or stay put.
Read the article →Context and a prompt can direct extraordinary computer work, but only when tools, permission, an action loop, and verification complete the equation.
Read the article →Apple's new AI partnerships are the right correction—but until they become broadly shipped capability, its closed-system advantage is becoming a speed risk.
Read the article →Why I track AI coding outcomes, API-equivalent cost, subscriptions, project load, and attention—and what the numbers cannot prove.
Read the article →A very concrete goal is a contract. Three sandboxed tests - each pre-registered with a way to fail - show what the letter of a goal delivers, and what it overrides.
Read the article →A human-readable way to design stages, decision gates, iterative voice calibration, and a publication loop that ends with evidence—not a green deploy alone.
Read the article →