Thoughts

Conversations in progress.

A working notebook for ideas I am developing, questions I am testing, and things that change how I see the work.

Some pieces begin with an observation. Others begin with a podcast, article, video, dataset, or conversation. AI may help challenge, connect, or shape the thinking — but the purpose is always the same: to look more carefully and arrive at a more useful judgment.

The evolution

Newest first

Thought 009 · Comic · 1 min · July 31, 2026

Imperialistic innovation

The model was broke. AI will accentuate the gap.

On the traditional IT playbook, and what it assumes about the ground it lands on.

A stick figure plants a large clay flag labelled THE SOLUTION on top of a hill, while three figures on the flat ground below carry a box, work at a bench, and haul a load, none of them looking up.

Thought 008 · Comic · 2 min · July 29, 2026

What have you stopped doing?

The gain isn’t the tool. It’s the step that stops happening.

On Google’s ATLAS report, Robert Solow, and a faster rewinder.

A drawing in three rows on one axis: renting, watching, rewinding and returning a tape; then a rewinder machine that shortens the rewind a little; then the rewind removed entirely, the line closing up and the reclaimed span marked in clay as the gain.

Thought 007 · Comic · 2 min · July 27, 2026

Easier feels easier.

The depth didn’t get harder. There are just fewer people down there.

A note on SQL, and on what you find on the way to the answer.

A section drawing: figures on a surface face a neat row of modeled tables, an empty ladder descends through an opening in the floor, and below one figure reaches into a heap of undocumented tables for one drawn in clay.

Thought 006 · Comic · 2 min · July 26, 2026

Taught to fish.

The point is not fewer questions. It’s better ones.

A field note from a program that trained departmental experts to do their own analytics.

Two panels: a person at an analytics counter asks for a report; beside them, a better-prepared person arrives with duplicate invoices already found and a sharper question.

Thought 005 · Read · 3 min · July 22, 2026

AI became a verb.

The letters reveal the work more clearly than the tool: AI assists, it acts, then people build what does the work.

Prompted by Netflix’s Q2 2026 shareholder letter and a wider scan of S&P 500 shareholder letters.

A four-stage timeline showing AI moving from something companies had, to something people used, to systems that act, and toward people building those systems.

Thought 004 · Comic · 3 min · July 20, 2026

Everything has a clock speed.

The risk isn’t the connection. It’s connection without deliberate design.

Prompted by Jason Clinton’s “Zero risk isn’t the job: a CISO’s guide to agentic AI.”

Five clocks stacked down a single dashed line, from years to seconds, with the fastest drawn in clay.

Thought 003 · Exchange · 4 min · July 19, 2026

Does AI enjoy AI-ing?

A bird cannot tell us it enjoys flying. An AI can tell us it enjoys thinking. The first answer may be better evidence.

Prompted by a study of whether galahs enjoy flying — and by the problem of asking a machine about its own experience.

A split drawing: a bird follows a looping flight path on one side while an AI response and cursor sit on the other.

Thought 002 · Interactive · 3 min · July 18, 2026

Propose McDonald’s.

When a room is stuck arguing about shadows, the way out is often not a better argument. It is a crude first version, placed on the table.

Prompted by Plato’s cave and a 2013 post called McDonald’s Theory.

A drawn scene: five figures turned toward a small glowing first version on a table, while the shadows on the wall behind them have shrunk to small ordinary objects.

Thought 001 · Essay · 4 min · July 18, 2026

Look at your fish.

Before adding another tool, another hire, or another layer of complexity, look harder at what is already on the table.

Prompted by Austin Kleon’s note on learning to see.

A rough napkin drawing showing the same fish in three progressively more detailed observations.