AI articles get their own address

The site was built so every entry keeps a flat top-level URL — /<slug>, never /<section>/<slug> — so a piece could be re-sectioned, or outlive its section, without its URL moving. That was right for a catch-all and stops being right now that Model Behavior is the one thing the site publishes journalism into. The articles want to read as a body of work at a shared address, so the section page can eventually be skinned as a publication rather than a filter.

The skinning is not this ticket. This is the plumbing only.

Nothing currently live moves. /where-world-leaders-stand-on-ai keeps its URL permanently, as do /my-git and /podcasts. No redirects are needed, because no URL changes — this is a going-forward rule.

Decided:

• A subdirectory is the opt-in. A file at src/content/entries/model-behavior/foo.mdx publishes at /model-behavior/foo; top-level files stay flat. No new frontmatter field, and the next exception is one more folder. Tickets already work this way.
• Home lists Model Behavior only, three at most, under "Latest AI" rather than "Lately".
• /model-behavior lists every article, uncapped. This is what makes orphaning impossible: an article always has a live link from the archive, so new pieces never have to chain back to old ones to keep them reachable. Linking them is still good writing — it is no longer load-bearing.
• The header says "Model Behavior", not "AI". Long on purpose: it is odd enough that someone who does not follow this stuff will click to find out.

Known consequence: with home filtered to Model Behavior, a future non-AI entry that is not a standing page would appear on no index at all. Nothing is in that position today. The fix when it happens is a second list on the home page, not a change to this.