For most of my career, two things limited what I could write: the hours I had and what I already knew. My articles reflected my own experience. That was honest, but it was narrow.
That limit has moved. Today I work with Claude as a research partner, a sparring partner and, some days, something close to a full team. Before I commit to a thesis, I can survey the research on it, find the work that disagrees with me, and test the argument from angles I would never have reached alone. My writing now draws on frontier knowledge, not only on my own memory.
Speed matters less than people think. What matters is that a CEO, CTO or investor who gives me ten minutes leaves with something they can use. AI lets me do the reading that respecting their time demands.
Here is what that looks like across my work.
Research and writing. I choose the question and the thesis. Claude helps me search widely, read deeply and find the weak points in my argument. I trace every figure back to its original paper, filing or dataset, and cut anything I can’t verify. It also helps me build the charts.
Final drafting. Once the argument is settled, Claude writes the final draft. Getting a piece release-ready used to be the most expensive step: hours of proofreading to catch spelling and grammar slips. Now my time goes into the thinking, and what reaches you is clean.
Engineering. Most of my engineering use of AI happens before any code is written. I use it for system design: laying out the options, stress-testing each against real constraints, and validating the choice before we commit to it. When it does write code on business-critical and compliance-heavy systems, it works within a strict discipline: every requirement traces to the code that implements it and the test that proves it.
Operations. I’m building agents for work that wears people down. One drafts customer-ready incident reports from internal engineering notes. Another troubleshoots new problems using years of past support tickets. The point is that engineers spend their time on what actually needs an engineer.
Strategy. When I’m weighing a product direction or a market, I ask Claude to argue the other side. A good sparring partner is worth more than a good assistant.
Learning. The partner that helps me reason about network architecture is also building me a structured guitar curriculum and helping with my homelab projects. Curiosity no longer has to wait for a teacher’s schedule.
Knowing where AI should run. Not everything belongs on a cloud model. Where data can’t leave, I run open models in-house. Deciding where AI runs is as much a part of using it well as deciding what to ask it.
What doesn’t change. I take full ownership of everything I do with AI, above all what I publish. What you read here represents my own thinking and understanding, completely. AI widens what I can know, but it doesn’t decide what I believe.
If you’re a builder, an executive or simply curious, my advice is this: stop treating AI as a faster search box. Give it your hardest question, ask it to prove you wrong, and keep the judgement for yourself. The leverage is real, and it grows the more you use it.