Know who actually wrote your next 1:1 agenda before you read it

A meeting-notes tool shipped a workflow on 10 August 2026 that lets a connected AI assistant create agenda items directly inside a shared meeting note, not just describe them in a chat window. ^source Nothing gets created unless a person asks for it, and whatever gets generated stays editable. That still changes what can end up in the shared record: a line in the agenda can now come from a request one person made to an assistant, worded however the assistant worded it, instead of from either person typing their own sentence.

That is a different shift than "AI in your 1:1" usually means. For a while, the interesting question was what an AI is allowed to see and summarize. Now it is what an AI is allowed to write.

Read access and write access are not the same risk

A tool that reads a meeting and produces a summary afterward is working from something that already happened. Whatever it gets wrong, both people can check it against what they remember saying or deciding. A tool that writes a line into the agenda before either person opens it is not reporting on the meeting. It is shaping what the meeting will be about, before either person has said a word.

That distinction is not just technical. The shared record in a recurring 1:1, the agenda, the notes, the list of what got decided, works best when both people can tell which lines came from a person and which came from a request to an assistant. Once a line's wording can come from something neither person typed, the record stops reading as a plain transcript of what the two of you said and starts reading as that plus whatever else got inserted.

An unattributed line breaks trust the way an unclaimed action item does

A 1:1 already has a familiar failure mode for commitments with no clear owner: the item gets remembered differently by each side, and by the next session neither person can point to what was actually agreed. A shared agenda line with unclear wording origin fails the same way, one step earlier. If the manager asks an assistant to draft the line, the direct report has no way to tell, just by reading the note, whether those are the manager's own words or the assistant's rendering of a quick prompt. The manager, in turn, cannot tell later whether the report read that specific wording as a firm ask or as a first draft worth pushing back on.

Once that uncertainty sets in, both people start treating the shared record the way you'd treat a group chat where you're not sure who is actually typing: still useful, no longer fully trusted.

The boundary we hold

When 1on1's AI works on a session, it reads: the prior sessions for that manager-report pair, the action items that closed or slipped, the trend in how the sessions have gone. The summary and the overall sentiment tag (positive, neutral, mixed, or concerning) it produces from that reading go to both people, same record, same words, both sides looking at the same thing. A separate, more detailed layer, a written-out read on engagement plus risk indicators and coaching prompts, goes into an addendum that only the series manager or an admin can open; the report's view never includes it, and that split is enforced with the same resource-level role checks documented for the rest of the product. ^source Either way, none of that output gets written into the agenda before a session starts. It exists to prepare the manager, not to draft what either person will say.

That is a category choice about what an assistant is for in a recurring 1:1. An assistant that reads the relationship to prepare a person is a different thing from an assistant that writes into what the two people say to each other, and it is worth knowing which one you are inviting in.

The test before you trust any shared 1:1 tool

Before you connect an AI assistant to a shared meeting note, whichever tool it is, ask one plain question: is this assistant allowed to write into the record, or only read it? That question matters more than whether it is helpful or whether it saves time.

If you want to see the difference in practice, run one manager and one direct report through 1on1's 7-day pilot: one reporting pair, a recurring 1:1, a week to watch how the shared record behaves. ^source Hold it to the same boundary described above: the AI can read context to prepare the manager, but nothing it produces gets written into the agenda before either person opens it. At the end, look back at the shared agenda lines from that week and check whether every one of them still has a person's name behind it. That answer tells you more about the tool than any feature list will.