#AI

Consumption pricing is suppressing the usage the frontier AI labs need most right now. From inside the enterprise, a simple fix: make the model cheap and sell the harness by the seat, at a price a company can actually predict.

The headlines say AI is coming to replace us. Watch the people who use it best and you see the opposite. They spend less time executing and more time thinking, judging, and directing the work. AI is handing back the most human parts of the job, and to everyone, not just the people at the top.

You can't measure an AI champions program by the tools it produces. The real signal shows up at the final meeting, when you hand each person a blank slide and watch what they choose to talk about.

Everyone watches the model. The harness around it, the thing that decides where it works and what it can touch, is the part that actually shapes your results. And it's the part you can build yourself.

I built a queue for the work I hand to Claude. It captures ideas the moment they hit, runs them one focused job at a time instead of as one overloaded prompt, and keeps a project's context alive across sessions. Here is why a queue beats a to-do list once the AI is doing the executing.

I built a personal company brain that I talk to every night. It works because it is built for one. Scaling it to a company runs straight into governance and access control, and that is the part still worth solving.

At a recent client kickoff, employees were weighing their own tasks in dollars for the first time. Consumption pricing gives finance a clean number, but it also teaches curious people to hesitate. A look at what the meter does once it reaches the floor, and a better way to handle it.

A single 45-minute run, 178 agents, and a laptop fan I had never heard before. What it takes to build a near-final slide deck on the jagged frontier of AI, and why someone has to be out on the edge at night to find out.

Three observations that show up at the start of every Gadoci engagement: the work is mostly about people, there's pent-up innovation waiting to be released, and the new shape of knowledge work is conducting, not typing. The notes I'd give you before you kick off your own AI program.

The Skills feature on ChatGPT and Claude is named wrong. In everyday English, skills live in you and references sit on a shelf. What actually shipped is a reference. Here is what that means in practice.

A description of what good AI work actually feels like, why it's held up across model eras and platforms, and why it's also the shape of knowledge work going forward.

A working through-line for what changed in how we use ChatGPT, why the term "context engineering" is doing real work, and the practical moves that separate people who get good output from people who don't. Includes a 30-minute coached exercise you can run on yourself.