A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: "What 102 AI Catalogs Say About The Identifier Change," Kin Lane, November 12, 2026, https://apievangelist.com/2026/11/12/what-102-ai-catalogs-say-about-the-identifier-change/
Marisol was a platform engineer at a mid-sized company that made scheduling software, and she read the post on a Thursday with her lunch going cold. Her team had published an AI catalog a few months earlier. She had a vague feeling that someone had told her the identifier format was settled.
It had not been settled. The post said it had been settled twice in two weeks, first with a kind suffix, then with a different separator once the first turned out to collide with URI fragments. Now it was a third form, using the URN's own colons. She felt a flicker of irritation. The author supported the new grammar, and she could see why. Publisher, then name, then an optional kind was tidy. But she had built against the earlier recommendation, and she had not asked to be a participant in a standards argument.
Then she reached the table, and the irritation turned into something closer to dread. Almost every published catalog used five segments, as hers did. The post said a five-segment identifier was still valid under the new grammar. It just meant something different. She read the next lines twice: "Nothing will reject it, and nothing will warn." And then, a sentence later, "A consumer will just read it wrong."
She opened her own catalog in another tab. Her entries put the word skill in the slot after the publisher, then the name of the skill. Under the new reading, every one of them would be named skill. She counted. There were six. The post described nine of those for another publisher, and she did not enjoy recognizing her own file in an example.
Her first instinct was to resent the whole thing. A grammar that lets old identifiers pass validation while quietly changing their meaning struck her as a design flaw, whatever the working group's reasons. She wrote that in the margin of her notes and underlined it. But the post did not pretend otherwise. It called the change cleaner and also laid out exactly what it broke. One line stayed with her: "People reached for a kind before the specification gave them one, and they put it in the opposite place." That was what she had done. Nobody had told her to, and she had needed a word for what each entry was, so she had written it first.
The fix was mechanical, the post said, and she believed it. Swap the two words when the first was a kind. Fold it into the name when it was a real namespace. She did not have to like the timing. The advice was to wait until 0.9 was tagged, and to move "before anyone builds an index that keys on your old identifiers".
She did not touch the catalog. The version was not final, and she had already been burned by one reversal that week, secondhand. Instead she created a plain text file called catalog-migration and listed her six entries. Beside each she wrote whether the slot after the publisher was a kind or a namespace. All six were kinds.
Then she added a reminder to check the specification's tag at the start of next week, and went back to her lunch.