Main idea

A session log fed at every conversation turn grows fast, and nobody rereads a text that lengthens without shape. The daily log template therefore imposes identical sections every day — context, objective, problems encountered, knowledge gained, work produced, decisions made, to follow up on, done — and the agent only has to add a line under the right one.

The template does two things at once, which is why it holds. It steers the writing: instead of "summarize the exchange", the instruction becomes "what, in this exchange, is a decision? a problem? a point to follow up on?", which rules out the flat summary. And it steers the rereading: you find a decision taken three weeks ago without rereading three weeks of journal, because you know which heading it sits under.

A less visible consequence: stable sections make deduplication possible. An agent that has to write under "Decisions made" can read what is already there and not rewrite it — something it couldn't do on free-form text.

Why it matters

This moves the work of structuring to the right place. The structure of material written as things go can't come afterwards: whoever rereads has neither the time nor the context to impose it. It has to be laid down before the first line, in the template.

It is also what lets you make a machine write without getting prose: the constrained format stands in for the editorial judgment it doesn't have.

Nuances and limits

Whatever fits in no section isn't written. An unclassifiable observation — an unease, an intuition about the direction of the project — disappears without a trace, and that is the exact price of rereadability.

And sections age: a template designed for one kind of work keeps partitioning work that has changed.

Open questions

• How often should a journal template be revised without losing comparability with older entries?