There's a small social ritual that has become normal in the last few years, and I don't think we've fully reckoned with how strange it is. A meeting starts. A gray tile appears in the participant grid: So-and-so's AI Notetaker has joined. Everyone notices. Nobody comments. And the conversation gets just a little more careful.

I've watched it happen in customer calls, in roadmap discussions, in candid one-on-ones. The bot joins, and something in the room changes. People hedge. The half-formed idea stays unsaid. The honest assessment of a partner, a deadline, a competitor gets softened into something safe for the record. The AI came to capture the conversation, and its presence made the conversation worth capturing a little less.

That's the paradox at the heart of the first generation of meeting AI: the more visible the AI, the less valuable the conversation it observes.

Why we built Earmark botless

When we started Earmark, we made a decision that shaped everything after it: no bot. Earmark captures conversations from your own device - across Zoom, Meet, Teams, Webex, phone calls, even in-person discussions - without sending a synthetic participant into anyone's meeting.

Some of the reasons are practical. A device-side approach means Earmark works wherever the conversation happens, on any platform, without asking your counterpart's IT department for anything, and without everyone in the room needing an account. A hallway conversation and a formal review get the same treatment. Your team doesn't change how it meets; the AI adapts to the meeting instead of the other way around.

But the deeper reason is about what a meeting is for. The best product conversations are messy, candid, and exploratory. That's where scope actually gets negotiated and rationale actually gets spoken. Anything that makes people perform for the record degrades the raw material. The AI should be invisible in the conversation and visible only in the results - the PRD, the tickets, the decision log that exist when you hang up.

Invisible AI, visible results. It started as a design principle. It's become the clearest way I know to describe the product.

Trust is a product behavior, not a policy page

The second reason botless matters is trust - and here I want to be precise, because this space is full of vague reassurance.

Product conversations are among the most sensitive discussions a company has. Unreleased roadmap. Competitive strategy. Candid customer feedback. Unfiltered engineering assessments of what's fragile. If you're going to let AI anywhere near those rooms, "trust us" is not an answer. The architecture has to be the answer.

For us, that means a few concrete commitments. Capture happens on your device rather than through a third-party guest in the call. Your conversations aren't used to train public models - what's said in your roadmap review stays out of anyone else's autocomplete. Retention is controllable: you decide what's kept and for how long, and your artifacts are yours, portable as plain Markdown you can take into any tool. And because there's no bot, there's no silent third party in the participant list that your counterparty never agreed to.

None of this is a claim that AI governance is solved - it isn't, and any vendor who says otherwise should worry you. Teams still owe their counterparts transparency about how conversations are captured, and security teams should still ask hard questions: where does audio go, what's stored, what trains what, who can see it. We built Earmark to have good answers to those questions, because we think the tools that win in serious organizations will be the ones that can survive that interrogation, not the ones that avoid it.

The results side of the bargain

Invisibility alone would be a gimmick. It matters because of what it protects: conversations candid enough to be worth converting into real work.

That's the "visible results" half. Earmark doesn't just capture the conversation - it turns it into the artifacts the meeting was supposed to produce, while the meeting is still happening. When ServiceTitan's product team evaluated Earmark, every participant reported significant weekly time savings, with two-thirds reporting five or more hours back per week - and over thirty days, the team grew organically from 7 active users to 33, from 291 meetings captured to over 1,100. Nobody mandated that adoption. People saw a colleague leave a meeting with the PRD already drafted and asked how.

That's the pattern I believe in for AI at work generally, beyond our product: the AI that wins won't be the AI you notice. It will be the AI that changed what existed when the conversation ended - the ticket that was ready, the decision that was recorded, the update that was sent - while the humans in the room did the one thing humans are for: thinking together, candidly, at full attention.

The bot in the participant grid was a first draft of meeting AI. It made the AI visible and the results thin. We're building the inverse.