Why I'm starting here
On the comfortable corner of the design space, why five commitments beat one mission statement, and what this column is actually for.
I’ve spent most of my career on the inside of this industry — first as a researcher, then for sixteen years inside one of the largest silicon organisations in the world. I’ve watched the design space for AI systems narrow, year over year, around one comfortable target: bigger transformers, optimised for data-centre throughput, measured by benchmarks that reward scale over judgment. It worked, for a while, because the workloads stayed roughly the same shape. They don’t anymore. Agentic AI is long-running and tool-using. Physical AI is power-constrained and safety-critical. And I don’t think we’re honestly exploring the design space those workloads actually need — I think we’ve gotten very good at exploring the same corner of it and calling that the whole map.
That’s the question behind the keynote I’m giving this week at ESSERC 2026, and it’s the question behind this journal too. I’m starting it because I don’t think the answer belongs only in a conference hall in Palma, or only inside Openchip. It belongs somewhere I can think out loud, in public, under my own name, and be corrected when I get it wrong.
I’ve organised what I write around five commitments, not one slogan: that systems should be built for and by humans, not around what’s easiest to benchmark; that they should be sustainable, because we have exactly one planet and the current trajectory already shows up in Ireland’s grid and in US electricity bills; that they should grow the next generation of talent instead of quietly assuming it will show up on its own; that they shouldn’t raise the cost of living through the back door of energy demand; and that they should close the gap between the handful of companies who can out-spend everyone else and the rest of the world, not widen it. Underneath all five sits a belief I hold plainly: none of this requires building smaller or less ambitious technology. It requires building the most advanced technology we can, on open standards, with the constraints treated as part of the design from the start rather than bolted on at the end.
What you’ll find here, roughly monthly: an argument, always dated and sourced, about one piece of that design space — sometimes about silicon, sometimes about the systems above it, sometimes about the people and policy around both. I’d rather publish something honest and unfinished than something polished and safe. If you disagree with a piece of it, that’s the point of writing it down.