
Start here
An assistant configured carefully once does not stay configured; vendors retire features, change defaults and rename settings on a rolling basis, and the only record of that is the release notes each vendor already publishes. A short monthly look at those notes, rather than a one-time setup, is what the documents themselves suggest is needed to keep a chosen configuration intact.
What the documents say
OpenAI's ChatGPT release notes, a running changelog, documented on 14 September 2026 that the company was 'retiring automatic switching from Instant to Thinking (reasoning)' for some plans and removing a related setting from the web interface, and on 11 September 2026 that custom GPTs were planned for retirement with a migration path to a different feature. The same page described a new default for a Health-data permission matching what most users, by OpenAI's own account, were already choosing — a stated default shifting to match observed behavior rather than a fresh explicit choice. Anthropic's release notes overview explains that the Claude Platform's notes cover the API, client SDKs and developer console, organized newest first, with a separate changelog for Claude Apps kept in its help center. Google's Gemini release notes page lists dated entries, including a Windows desktop launch and a macOS voice-dictation feature, both announced in the weeks before this guide's preparation date. All three describe change happening on an ongoing, dated basis rather than in occasional large jumps.
Check this
This is an editorial routine built from the three pages above. Once a month, open the release notes for each assistant in regular use and read only the entries since the last check. Note anything described as a retired feature, a changed default or a renamed setting, since those are the categories the OpenAI notes above actually contained in a single week. Then re-open the settings pages identified elsewhere in this guide and confirm the chosen configuration is still in place, since a vendor changing a default does not necessarily notify an existing user directly.
What holds and what fails
This routine holds for catching changes a vendor has documented; it does not catch undocumented or silent changes, and none of the three pages claims completeness. It also assumes the check stays short, reading only new entries, since a full changelog each month is a heavier habit than most readers will sustain. This is an editorial routine assembled from three vendors' own release-note pages, not a guarantee a configuration survives unattended between checks.
- Set a recurring monthly reminder to skim each assistant's release notes, not only its settings page.
- Watch specifically for words like 'retiring,' 'default' and 'removing' in each month's entries.
- Re-check a deliberately chosen setting whenever its related feature appears in the notes.
The assistant that answers a question in October is rarely configured exactly as it was in September, and a vendor's own notes are where that difference gets written down.
Sources & reading trail
Dated changelog entries documenting a retired auto-switching setting and planned custom GPT retirement.
Source published: Not established · Retrieved: 16 September 2026
Describes the scope and chronological structure of Claude Platform release notes and the separate Apps changelog.
Source published: Not established · Retrieved: 16 September 2026
Lists dated Gemini Apps feature releases in the weeks leading up to this guide's preparation date.
Source published: Not established · Retrieved: 16 September 2026
Documentation, regulator guidance and studies establish the record; the checks and the boundary are AI Use Field Guide editorial analysis. This retrospective draft does not imply the site published on the event date.