Commander's Intent for AI Agents
The Army taught me to give intent, not step-by-step orders, and trust the people closest to the problem. It turns out that is also the secret to getting great work out of AI.
Notes from running a fintech with AI — plus the occasional piece on investing and leadership.
Going deeper? Technical build logs live in The Lab →The Army taught me to give intent, not step-by-step orders, and trust the people closest to the problem. It turns out that is also the secret to getting great work out of AI.
People are asking AI instead of Googling. That quietly rewrites the rules of being found. Here's how I evaluated Profound versus Athena for work — and why we went with Athena.
A routine visit to a specialist turned into a billing mess and a denied claim. Most people give up at that point. AI is the reason I didn't — and why I filed a formal complaint with the state.
I ran the major AI meeting-notes tools through real work — including a head-to-head with the Gemini note-taker built into Google Meet. One quietly won. Here is why Granola earned the spot.
Claude Opus 4.8 is a real step up and my everyday workhorse. But the first time I handed a sprawling, multi-step problem to Fable — a long-horizon model — I saw where this is actually going.
My tax preparers couldn't get me my numbers in time, so I uploaded everything to Claude and estimated my own taxes. When the pros finished, they were super close — and the planning paid off all year.
I built the first version of this site with Claude in February. Three months later I rebuilt it — same tool, new design capabilities, totally different result. A walk-through of what Opus 4.7 and the Claude Code desktop app actually changed, with before/after screenshots.
I have a CPA, a financial advisor, and a doctor. I am not replacing any of them. But the way I interact with all of them has fundamentally changed this year — and AI is the reason.
I built a personalized Claude quick-start guide for a friend, hosted it on my website, and then turned the whole thing into a database-backed content platform with admin controls — all in a single session. The speed of iteration is getting absurd.
I shipped five features in a week using AI. Then I asked it to audit what I built — and it found a hole I never would have caught on my own. The technical gap is closing. What is opening in its place matters more.
A contact form that logs leads to a database. A comment section for readers. A private admin dashboard. All built without writing code — and what it reveals about the new economics of building.
A non-technical operator goes from blank screen to live custom website in one afternoon — and discovers that the bottleneck was never technical skill. It was something else entirely.
The executives pulling ahead are not waiting for IT to roll out an approved solution. They are running their own experiments — and building a cognitive advantage that compounds every single week.