Friendly, practical writing on doing real work with AI — where it earns its keep, when to trust it, and how to keep the capability you build. No hype, no jargon.
A well-prepared client came to me with a list of painful processes and real money attached, and I passed. They wanted automation laid on top of the company exactly as it already ran, and that would have worked fine and changed nothing.
I hired someone who doesn't build agents, write prompts or touch our automation stack. She sits with people and helps them figure out what changes now that they have an AI colleague, and that turns out to be where almost all the value is.
Jonathan and I filed a provisional patent this week using four instances of Claude passing work around in a circle. The one that mattered was the one whose only job was to tear our work apart.
Three teams looked at the inventory system we're building for Wine to Water and got stuck on one wrong number. That wrong number was the whole point. Claude turned the feedback into 27 buildable changes and contributed nothing to generating it.
A VP of Marketing spent the first twenty minutes of every meeting extracting information she couldn't see. We put every channel in one place, pointed Claude at it, and the whole shape of the conversation changed. The Top Cup story.
We started with the sales team because that is where the appetite was. The work was good — and it went nowhere. What I learned about why changes in how a company works only ever flow down, never up.
Have a piece of work you think AI should be doing?
Book a call
