# Field notes — week of October 19, 2026 > Source: > Published: 2026-10-26 00:00:00+00:00 # Field notes — week of October 19, 2026 What changed in AI-support and AI-localization tooling this week. Every item was checked against its source before it entered this list. - **Anthropic** -[launched Claude Models on Google Cloud (Anthropic managed)](https://platform.claude.com/docs/en/release-notes/overview) (Oct 22), running the Claude API through a customer's own Google Cloud project with access controlled by Google Cloud IAM and billing through Google Cloud Marketplace; two days earlier it shipped user profiles, letting API callers tell Anthropic which end customer a request is for so enforcement actions - restrictions or bans - can target just that customer (Oct 20). - **OpenAI** -[took the Decisions API to general availability on gpt-6-luna](https://developers.openai.com/api/docs/changelog) (Oct 22), adding prompt-cache discounts, image support in question instructions, a structured-data`input_json` option and up to 400 questions per request; two days earlier it added 12-hour prompt caching to the Responses API, keeping shared instructions and tools cached longer at a lower cached-input price, with cache writes billed at 2.5x the input-token rate (Oct 20). - **Intercom** -[added Flexible Snooze Options](https://www.intercom.com/changes/en/161312-flexible-snooze-options) (Oct 23), letting admins define custom snooze durations, group them into lists and assign each list to specific teams; two days earlier it let teammates[follow a conversation or ticket](https://www.intercom.com/changes/en/160981-follow-conversations-and-tickets) to get the assignee's notifications without being assigned to it (Oct 21). - **Sierra** -[published 𝜏-i18n](https://sierra.ai/blog/tau-i18n) (Oct 21), extending its voice-agent evaluation to Spanish, Brazilian Portuguese, Hindi, Korean and Mandarin, measuring task completion alongside interaction and generation quality; the next day it published[𝜏-Entity](https://sierra.ai/blog/tau-entity-can-voice-agents-get-the-details-right) (Oct 22), a benchmark finding that voice agents often struggle to capture details like names and phone numbers correctly, even on simple single-field calls. - **Decagon** -[introduced Experiment Hub](https://decagon.ai/blog/experiment-hub) (Oct 22), an experimentation platform that lets teams test one change to their AI agent while other updates keep shipping, then analyzes results and helps pick the winning version. - **WPML** -[shipped WPML 5.2](https://wpml.org/announcements/2026/10/wpml-5-2/) (Oct 21), which stops page saves from overwriting translations that didn't change, rebuilds the Translation Dashboard to work with any number of languages, pauses automatic translation with a billing prompt when payment is blocked, and includes 565 other fixes.