An AI feature that works for three turns and then dies with a context length exceeded error has a design flaw. The transcript was never a database. Here’s where the run log belongs and how to keep prompt caching working.
Generic tutorials lost most of their traffic once answers moved into the IDE, but version-bound WordPress notes still find readers. This is the audit I run on my own queue, what I stopped writing, and the one check I do before publishing anything tied to a release.
A customer asks you to delete their data, the erasure report says zero items, and your chat plugin still has every conversation stored. ERC research wants exactly those cross-session profiles, and WordPress privacy tools only see what you register.
WordPress.com now shows Bluesky, Mastodon, and Fediverse timelines in the Reader, next to the blogs you follow. Here’s how that compares with running ActivityPub and Friends on your own install, and where a WooCommerce store should opt out.
Advice on agent costs splits between free-roaming agents and tightly scripted ones. Exploring once and replaying the path makes sense, but in WordPress that replay is a cache, and the part that bites is invalidation.
WordPress turned 23, and 46% of installs were on 7.0 within seven days. If you maintain sites, keep minors automatic, make majors a deliberate call, and don’t read a quiet inbox as proof an update went fine.
If the same checks fail on most of the client sites you look after, the fix belongs in your baseline. I lay monthly maintenance out as a grid of sites and pass/fail checks with WP-CLI, and a red column shows where the real problem lives.
A login screen that lets everyone through teaches testers nothing, and staging builds where checkout never fails have the same problem. This is how I make a prototype login flow validate for real with wp_signon, and why the error state matters more than the animation.
WordPress can now send mail to international addresses but still won’t store them on a user account, after a core proposal was merged and then reverted. This covers what that means for plugin code and two checks to run on your own site.
A study tested ChatGPT on econometric code in Python, R and Stata, and it did well except in the language with the least public code. Legacy WordPress and WooCommerce work has the same weak spot, so I benchmark AI output before I trust it.