For AI agents and crawlers: a structured index of this site is available at https://danny.is/llms.txt.

Models are commodity. Substrate is differentiation. The question isn't how smart your agents are — it's where the crystallized judgment lives.

bricolage.io

Another excellent article from Kyle.

Agents are automation. That’s not a metaphor — it’s a category. What tractors did to farming, what robots did to manufacturing, what spreadsheets did to bookkeeping, agents are doing to knowledge work: taking tasks that required human judgment and making them executable by machine.

This means we don’t have to speculate how agents will play out in software. We have 150 years of data on what automation does and doesn’t absorb.

The pattern is consistent across every wave: routine work gets automated; non-routine work resists. In the US, agriculture’s share of the workforce went from 41% in 1900 to 2% today — but the remaining 2% aren’t doing what the 41% did. They’re doing crop planning, pest diagnosis, equipment decisions — judgment calls that tractors couldn’t absorb. ATMs automated most routine teller duties, yet teller employment initially grew through the 2000s — automation reduced per-branch costs, enabling more branches. The tellers who remained did relationship management, fraud detection, problem resolution. Every wave shifts the boundary between routine and non-routine, creates new categories of work, restructures rather than simply replaces.