Writing about AI executive assistants, building systems that work, and what happens when you stop fighting your own wiring.
We spent four days chasing a bug with increasingly clever fixes. The solution was making the system refuse to accept bad input in the first place.
Claude Code v2.1.142 deprecated TodoWrite. The replacement works -- but fails in three invisible ways if you have custom workflows. Here's what broke and how we caught it.
Dan Hollick uses AI to build illustration tools, not to make illustrations. That's the version I keep coming back to: AI that speeds up the manual work without giving up control or quality.
AI-generated content feels offensive even when it's competent. I think I know why.
The instinct is to give Claude everything. But there's a point where more context makes the answers worse, not better. Here's the mental model that explains why.
Claude Code is powerful out of the box - but it assumes you're a software engineer. Here's how I used a D&D-style character sheet to make it match my work and communication style.
AI is a thinking partner, not a replacement for thinking. The difference determines whether it makes you sharper or replaceable.
I gave my AI assistant access to my inbox, my DNS, my membership platform, and a shell. Every one of those is an unlocked door. Here's the security system.
My AI assistant had been extracting memories from every conversation for months. Thousands of them. Then I looked at the data and realized the system was drowning in noise.
Five months of tangled expenses, four trips with different splits, and two failed spreadsheet attempts - solved in 20 minutes of back-and-forth with an AI assistant.
How a simple question about finding old conversations turned into a unified search system - and why the design process mattered more than the code.
Five days into building my AI exec assistant, I asked it to write down what we'd done so far.
AI handles everything that isn't the writing. Here's the philosophy behind the newsletter workflow that cut 4 hours down to 30 minutes.
How deterministic scheduling, jitter, idempotency, and queue-based processing keep an AI executive assistant running around the clock without stepping on its own feet.
I shared an essay about another AI system. What came back wasn't a summary - it was a reflection. Then they asked to write about it.
Proven techniques for AI adoption that actually stick - momentum, safety, engagement cadence, and why behavior change matters more than the model.
I spent 20 years running businesses before AI could actually help. Here's why I stopped waiting for the right product and started building.
Most AI tools ask for your data and act on your behalf. Here's why I built a system where you stay in the loop on every decision that matters.
How I went from not writing code in 15 years to building an AI executive assistant - and the moment I realized Claude Code wasn't really a coding tool.
Most Claude Code skills are static instructions. Mine improve through structured feedback loops - capturing edge cases, updating their own docs, and compounding over time.
Claude Code starts every session smart. Your job is to keep it that way. Here's how I manage context across multiple businesses without losing my mind.
Open loops - tasks waiting on someone else - used to consume my entire working memory. Building a system I actually trust changed everything.
Every Claude Code session generates knowledge that vanishes when the session ends. I built a system that catches it automatically - and the compounding effect changed how I work.
I built an AI system while my business partner was in Japan. When he got back, he wasn't impressed. He was concerned.
Plain text files, organized well, gave my AI assistant a memory system that surfaces recommendations I'd never think to ask for.
I spent 15 years looking for the right productivity tool. Then I stopped looking and built the assistant I actually needed.
No spam. No "AI newsletter." Just an email when it's time.