OPINION
Tools move from called to made
One testable call per piece: if it is wrong, it should be possible to name the evidence that refutes it.
Issue 02 · Agents and tools
Agents no longer just call tools; they start forging their own. That rewrites our default assumptions about boundaries.
More essays in the edit: a few pieces per issue, never filler.
OPINION
One testable call per piece: if it is wrong, it should be possible to name the evidence that refutes it.
TOOLS
Reusable procedures and checklists. Short, but every step can be run; no conclusions without process.
NOTES
Only changes that shift judgement: model capability, tool interfaces, regulation. Not a news ticker.
The parts this publication is built from
TOOL
Markdown in, validated index and prerendered HTML out; out-of-contract fields fail the build.
Live
NOTE
Links only when translationOf is reciprocal; a one-sided pair is an orphan error, never path inference.
Live
TOOL
Canonical, hreflang with x-default, and OG tags generated per article.
Live
TOOL
sitemap.xml and robots.txt written at build time; drafts never reach the output.
Live
TOOL
data-theme plus localStorage, written before first paint and kept across reloads.
Live
NOTE
Heading anchors generated at build time; TOC, reading progress and citations share one source.
In progress
TOOL
Two families, two weights, self-hosted and subset to real usage; missing glyphs fail the build.
Planned
NOTE
Section and date archives plus site search, after the bilingual library stabilises.
Planned
VIEWS
Agents no longer just call tools; they start forging their own. That rewrites our default assumptions about boundaries.