All Posts

OpenAI's math catalogue is progress arriving in the hardest knowledge work. If it can happen there, every industry that still treats its smartest people as the pace is late.

TypeSafe named its model after Kahneman's fast thinking and Jevons's cheap coal. Both bets say something about where agent decisions are headed.

Anywhere a human gave input to push an output closer to what they actually wanted, that interaction contains a precise description of their standards. The trick is to stop letting it evaporate.

Every task you delegate to an agent is a primitive. Primitives combine. The real value is not the task you just completed, it is the capability you just added to every future workflow.

The engineering backlog is a tax on organizational creativity. Give everyone—sales, CS, product, design—the same tools engineers have, in a safe sandbox, and watch what emerges.

The DORA 2026 report shows incidents rising sharply alongside AI-driven throughput gains. The dip is not turbulence to endure. It is a diagnostic about which systems in your organization need redesign.

Operating at exponential pace is not about working harder. It is about restructuring how work flows: parallel streams, shallow planning, closed loops, and agents doing the building.

A stronger interview loop starts with an AI-allowed take-home and then tests judgment, depth, and adaptability by extending the same problem onsite.

Karpathy's AutoResearch loop can be extended from model experiments to product strategy, backlog generation, and continuous feature evolution.

AI agents are replacing collections of consumer apps with software built for your specific life—and the economics finally make this possible for everyone.

How a PRD, seven RFCs, and a bash orchestrator turned specifications into working code without a human touching an editor.

Recover the source of truth from the product, enforce it like coverage, and keep different artifacts in sync.

Why clear intent and executable specs are moving to the center of software engineering.

How AI-first development changes problem-solving, why system experts are always available, and what this means for engineering teams

The shift from perfect code to good-enough code, and why build times and release processes are the real bottlenecks in modern software development

The cultural divide emerging in engineering teams between AI adopters and resisters, and how it's reshaping team dynamics, knowledge sharing, and career progression

As AI transforms software development, the traits that separate great engineers from good ones are evolving. Here's what to look for when building your team.

How AI is accelerating the evolution of senior IC roles and blurring the line between individual contributors and managers

My first blog post introducing this new site