The scarcest thing in any meeting is attention. People are listening, thinking, reacting, deciding, reading the room, and trying to work out what actually matters - all at once. Yet a lot of workplace software treats that attention as free: click this, tag that moment, pick a template, confirm the action, fix the note while the conversation keeps moving. Building Earmark taught us how backwards that is. If someone has to operate the AI during the meeting, they're not fully in the meeting - and the meeting is the one place you least want to pull them out of.

That became a real design principle for us: the meeting isn't where you make the user do more work. It's where you let them stay present. The value should arrive afterward - the spec draft, the engineering handoff, the decision log, the customer follow-up, the project update, the open questions, the tasks waiting for review - without the user having to steer the tool the whole way to get there.

Bad AI demands attention exactly when attention is most valuable. Good AI protects it, and returns the leverage afterward.

This matters most for product and engineering teams, because the best parts of a meeting tend to be the subtle ones: a customer's hesitation, a constraint engineering raises in passing, a quiet disagreement between design and product, a tradeoff that only comes into focus ten minutes into the discussion. If the PM is busy managing a tool, they can miss the very thing the tool exists to help with. The tool should adapt to the meeting, not force the meeting to adapt to the tool.

None of this means AI should vanish entirely. Review, judgment, and approval all still matter - their moment is just after the conversation has produced context, not in the middle of the team creating it. The best version is the one you forget is running, until you open the output later and find it captured the work better than you would have. Not loud AI, not performative AI, not a co-pilot that constantly asks to be flown - a quiet system that understands enough to hand back useful first drafts once the meeting ends.

The more we build Earmark, the more we treat attention as one of the most important surfaces in the product. If AI is going to spend it during the work, it had better earn that cost - and most of the time, the better move is to stay out of the way. Let people think, listen, and decide. Let them be fully in the conversation. Then help turn that conversation into the work they need next. The best AI meeting experience isn't the one people notice most during the meeting. It's the one they trust most after it.