Writing about engineering leadership, AI infrastructure, and what it takes to operate at executive scope.
Most people think the jump is about scope — more people, more budget, more surface area. That's true but it's not the hardest part. The hardest part is the shift from managing execution to owning outcomes you can't directly control. Here's what I've seen separate the people who make the transition cleanly from those who struggle.
Read essay →Everyone's racing to buy more GPUs. The constraint that actually limits most hyperscale training runs is something far less glamorous: interconnect topology, memory bandwidth, and the engineering discipline to keep utilization high. A look at what the frontier actually looks like from the inside.
Read essay →Output is what your team ships. Outcomes are what the business achieves because of it. This distinction sounds obvious, but almost every promotion failure I've witnessed — from staff to principal, from director to VP — comes down to a person who kept optimizing for the wrong one.
Read essay →The graveyard of internal developer platforms is enormous. Teams pour years into beautiful APIs that nobody adopts at scale. The difference between the platforms that win and the ones that fade isn't technical quality — it's a set of forces most platform engineers never learn to see.
Read essay →The industry spent five years obsessing over training clusters. Inference — actually deploying intelligence at scale, with latency and cost constraints — is a different problem entirely. Here's where the real engineering challenges are, and where the next decade of compute architecture is being decided.
Read essay →