Agents & Agency

The Lab · Exhibit J

Three essays you can play with

Each figure turns the argument of one essay into a small model you can push on. They make the assumptions visible. None of them is a forecast.

This page has an agent

It's the only page on the site that does. Every so often it will offer to do something you didn't ask for. You decide. That's the argument of Unprompted, acted out.

No unprompted offers yet this visit.

Fig. 1

Who moves first?

Read Unprompted →
  1. 01Notice a needYou
  2. 02Choose the goalYou
  3. 03Plan the workYou
  4. 04Do the workAI, when asked

Hammer. You want a playhouse for your kid. You pick the design, buy the lumber, and drive the nails with a nail gun. The tool does what you aim it at.

Decisions that stayed yours3 of 3
Fig. 1. Based on my July 2026 essay Unprompted. Chat tools work like a hammer. Agentic tools work more like a contractor. The next step is agents that choose the goal themselves and act before anyone asks. The essay asks what that does to human agency.
Fig. 2

Solvent on paper, short when it counts

Read the essay →

How fast the profession hands entry-level work to models.

Deliberate spending on the slow way: juniors doing hard things next to seniors.

Fig. 2. A toy model from my January 2026 essay The Liquidity Crisis of Expertise, which argues a profession can run out of usable judgment the way a bank runs out of cash. Seniors retire, juniors learn by doing the work AI now absorbs, and captured expertise takes about a decade to mature. It illustrates the argument and forecasts nothing. The full simulator in the Lab adds a third slider.
Fig. 3

Challenger or sycophant?

Read the essay →

We should replace our intern program with AI agents.

Sycophant agent

Smart move. Agents work weekends and never need parking.

Feels great and teaches you nothing.
Challenger agent

Where will your senior people come from in ten years if nobody starts at the bottom?

Mildly annoying, and it shows you where the plan breaks.
Fig. 3. From my October 2025 essay The Rise of the Agentic Class. It argues the biggest design choice in AI is whether agents challenge our thinking or flatter it, and that the answer will sort people into a new digital divide. Same question, two agents.