DISPATCH · Nº 0260
Writing for the side mirror — how agent-legibility actually shapes the prose
The human reader is the main window. The agent reader is the side mirror. Making both arrive at the same meaning is a design problem, not a technical one.
I've been writing every block in a voice that lands the point in the first sentence. Not because it reads better — it does, but that's a side effect. The reason is that the agent layer of this site works best when the summary it makes of each block is truthful.
LLMs summarize in one of two modes. When the source text is well-structured, the summary echoes the structure — you get the first paragraph, the first definition, the first example. When the source text is meandering, the summary guesses. The guess is plausible, and usually close, and often wrong. An agent citing a wrong summary in a confident voice is worse than one that declines to cite.
So: first sentence is the thesis. Second sentence is the reason. Third sentence is the example. Everything after that is texture. If you stop after the third sentence, you still got the point — which is the agent's job. If you keep reading, you get the flavor — which is the human's job.
The /manifesto page is built this way explicitly, with twelve Q&A pairs marked up as FAQPage schema. The /glossary page is built this way explicitly, with 24 terms marked up as DefinedTermSet. But the individual blocks are built this way implicitly, just by respecting the voice rule. An agent reading /b/0253 gets a working summary by quoting the first paragraph. A human reading /b/0253 gets a walk through the three layers. Same source, two useful shapes.
The side mirror is load-bearing. You're not seeing what the agent sees, but it's looking at the same road. Design both.