An AI Assistant for Your 1:1s — What It Can and Can't Take Off You

Do the arithmetic on your recurring 1:1s before you read anything else.

Eight of them a week at thirty minutes each is four hours. Add the five minutes before each one where you open the doc and remember what you talked about last time, and the ten after where you write down what you promised — that's another two. Six hours a week, which is roughly fifteen percent of a working week, spent on conversations that mostly follow the same shape every time.

None of that is wasted. Recurring 1:1s are how you find out things are on fire before they're on fire. The problem isn't that they're low value; it's that they're fixed cost, and the cost scales with your team while your week doesn't.

What "AI assistant for 1:1s" usually means, and what I mean

Almost everything sold under this name records the meeting you were already in and writes it up. That's a notetaker. It makes the hour more useful and it doesn't give you the hour back.

What I'm describing takes the call. An AI persona anchored to you meets the other person, has the actual conversation, and hands you the result. The hour comes back.

Two things that matters most to be clear about, because they're the questions everyone asks in the first minute:

It doesn't join your Zoom or Teams call as a bot. The meeting happens on Waterr, at your own link — waterr.ai/{username}/{scenario-slug}, or a scheduled invite with a specific slot. This isn't an attendee added to a calendar invite for a meeting that was happening anyway. It's the meeting itself, relocated.

It's about calls, not messages. This doesn't triage your Slack or answer your email. It handles conversations that would otherwise have been a scheduled call.

What it actually does on the call

The persona runs a scenario you configured — what the conversation is for, how it should behave, what it's trying to find out. During the call it holds that position rather than drifting into generic helpfulness at minute twenty, pursues the goals inside the time budget, and can see: expressions, posture and environment through the camera, and documents, spreadsheets, application interfaces, code, and diagrams through a shared screen.

It can also remember. Participant memory captures what the person shared — role, company, what they're working on — plus stated preferences and the loose threads they didn't finish, and injects that as ambient context before the next call rather than reciting it back at them like a CRM record. Two limits worth knowing up front: it's off by default and enabled per scenario, and in v1 only signed-in participants are remembered, so someone joining as a guest through a public link isn't.

When the call ends, the transcript, the analysis, and the recording come back to you. The analysis scores the conversation against the goals you set — each with a name, a description of what good looks like, and scoring instructions — on a scale, typically 1 to 10, with written feedback and transcript highlights linked to the specific goal each supports.

You can also add context per send without touching the underlying scenario. Same weekly check-in scenario, but this week: dig into why the migration slipped.

What it isn't

Being precise about this is more useful than another paragraph of benefits.

Not a bot in your existing meetings. Nothing joins your Zoom, Meet, or Teams calls. The meeting runs on Waterr.

Not a message triager. Slack and email are a different problem and this isn't it.

Not a calendar rearranger. It doesn't find slots or move things around. It removes the need for the slot.

Not for the conversations that matter most. More on that below, because it's the important part.

The 1:1s worth handing over

Sorted by how well they actually work, from what I've seen:

Recurring status check-ins that produce a report. The weekly sync where someone walks through what shipped, what's blocked, and what's next. It's structured, the output is informational, and it happens fifty times a year. This is the strongest case by a distance.

Skip-levels you keep cancelling. The one that's been rescheduled three times because your week keeps winning. A conversation that actually happens with an AI beats the fourth cancellation of one with you — as long as the person knows the real version is still available and is genuinely easy to get.

Vendor and partner check-ins. Regular, low-stakes, informational. Nobody's career is affected by the format.

Pre-conversation gathering. The half-hour of context-collection before a decision meeting. Have the agent gather it, arrive with it already in hand, and have the actual discussion with a human.

Customer check-ins during a crunch. Not the QBR. The "how's it going, anything broken" call that gets skipped during a launch week and then becomes a churn conversation two months later.

The 1:1s you shouldn't

Performance conversations. Anything with bad news. Anything where the person is struggling. Anything where the relationship is the point.

That's not a limitation I expect to engineer away, and I don't want to. If someone's had a rough quarter and gets your AI instead of you, you haven't saved thirty minutes — you've sent a message, and it's a worse message than the meeting would have been.

The judgement is straightforward: if the conversation's value is in the information, delegate it. If its value is in the fact that you showed up, don't.

The objection everyone has

"Isn't this rude?"

Sometimes, yes. It's rude in exactly the cases above, and treating it as a universal rudeness is a mistake in the same way that treating every calendar invite as a compliment is.

Two things make the difference in practice. Transparency — there's a consent modal before the device check carrying whatever disclosure you write, and typically stating that the session is recorded, transcribed, and reviewed. Nobody is being tricked; they'd know within one exchange anyway. And the alternative — the honest comparison isn't agent versus you. It's agent versus the meeting that got cancelled twice, or the Slack thread that ran for four days, or the update that never happened because nobody had a free half hour.

Measured against a great 1:1 with you, this loses. Measured against the meeting that didn't happen, it wins comfortably. Most recurring 1:1s are competing with the second one.

Frequently asked questions

Can an AI actually run my 1:1s? For structured, recurring, information-shaped ones, yes. The persona runs your scenario, holds its position for the length of the call, sees camera and screen share, and returns the transcript, a scored analysis, and the recording. For conversations where your presence is the point, no — and you shouldn't try.

Does it join my Zoom or Teams meetings? No. The meeting runs on Waterr at your link, either open or as a scheduled invite with a specific window. Nothing gets added to your existing calls.

Does the other person know it's an AI? Yes. There's a consent step before the device check with your disclosure message, and they see the persona's name and avatar on the welcome page before starting.

Does it handle my Slack messages too? No. This is about calls and scheduled conversations. Chat is a separate problem.

Will it remember what we discussed last time? If participant memory is enabled on that scenario and the person is signed in, yes — facts they shared, their preferences, and threads they left open, carried in as background context. It's off by default, so switch it on for anything recurring.

What if they'd rather talk to me? Then talk to them. Keep the path to a real conversation short and obvious. An agent that's a substitute for an unavailable half hour works; one that's a wall around you doesn't, and people work out which it is very quickly.

More on the mechanic in send an agent, on the arithmetic in the coordination tax, and on letting people self-schedule in the AI booking page.