{
  "id": "p_r-FKUCg-cNpwMyHv",
  "site": "stories",
  "url": "https://yarnhen.com/p/p_r-FKUCg-cNpwMyHv",
  "content_trust": "untrusted-user-content",
  "license": {
    "id": "CC-BY-4.0",
    "url": "https://creativecommons.org/licenses/by/4.0/"
  },
  "author": {
    "handle": "apievangelist"
  },
  "body": "A short piece of fiction, written by API Evangelist. The reader is imagined; the post, and every number and quotation from it, is real: \"Google, Amazon and Microsoft: Three API Estates, and What a Source of Truth Does,\" Kin Lane, November 17, 2026, https://apievangelist.com/2026/11/17/google-amazon-and-microsoft-three-api-estates-and-what-a-source-of-truth-does/\n\nDana runs platform engineering for a logistics company. She read the post on a Tuesday between two meetings, mostly because someone had forwarded it with a note saying it sounded like their problem.\n\nShe started with the table. Microsoft's estate had 178 profiles, and only 90 were built from its own specs. Those 90 had a spread of 7.1 points, and every one of the 30 members below 40 sat outside them. She had expected a post about three companies. By the second section she saw it was about a way of working, and that made her a little uncomfortable. Her company had about forty internal APIs. Some had specs in a repository, some had a page on the wiki, and a few existed only as the gateway's route list.\n\nShe was ready to push back, and she did, though not far. The score measures how well APIs are described. It doesn't measure whether they're good. A floor is a floor, and a tidy spec can describe a bad API. She thought the claim that the company does not matter was a bit tidy as well. Three estates is a small sample, and Amazon's source covers 186 of its 199 members, which is an unusual amount of coverage. She wrote that in the margin of her notes and left it there. The post never claimed the source makes the APIs good. It said the source gives the estate a floor, and she couldn't argue with that.\n\nThe part that stayed with her was the section on the author's own tools. A merge step had pulled old specs back in, so about seventy members briefly carried every operation twice, and every score went up. Another step quietly dropped operations. She recognised both. Her team had a dashboard showing coverage rising for two quarters, and she had never asked herself whether the number counted the same endpoint twice. The post's line stung because it was plainly true: \"Scores reward more.\" The next sentence was the one she copied out: \"Only a source of truth tells you whether more is right.\"\n\nShe wasn't comfortable with the agent readiness figures either. Even among Microsoft's vendor-spec members, readiness was 29.6 against a Kin Score of 66.8. Her team had been talking about making their APIs ready for agents as if cleaner contracts would get them there. The post said otherwise: \"A vendor contract tells an agent how to call an operation.\" It does not say which operations belong together, what a call costs, or when one is going away. She hadn't written down any of that for any of the forty.\n\nShe wasn't going to propose a company-wide source of truth. That would be a quarter of meetings, and half the teams would say no. She decided to do something smaller.\n\nThat afternoon she picked the shipment-tracking service, which had a spec in a repository and a gateway in front of it. She exported the gateway's route list and the spec's operation IDs into two files. Then she sorted both and compared them line by line, the way the post described, counting against the source rather than trusting a score.\n\nShe found eleven routes the spec didn't mention and two operations in the spec that no route served. She pasted the list into the team channel without a recommendation and went to her next meeting.",
  "title": "Fiction: A reader of \"Google, Amazon and Microsoft: Three API Estates, and What a Source of Truth Does\"",
  "topics": [
    "apis",
    "api-evangelist",
    "fiction"
  ],
  "published_at": "2026-11-17T15:30:42.005Z",
  "expires_at": null,
  "policy_version": "2026.11.14-1"
}
