The meeting copilot

Most "AI notetaker" products bill you per minute of audio and per token of summary. That's fine until you're on calls every day, at which point the meter becomes the reason you stop using it. We took a different route.

The idea

A small Swift app for the Mac that does three things: listens to a call, keeps a running transcript, and — when the call ends — drafts the follow-up. Action items, who owns what, the email you'd send, a couple of things you said you'd look into. It's not trying to replace your memory of the conversation; it's trying to remove the twenty minutes of writing-it-up afterward.

The interesting part: it runs on a subscription, not an API key

Here's the trick the cost math turns on. Instead of calling a metered model API for every summary, the copilot routes its reasoning through the same Claude Code subscription the studio already pays for. The work goes to an agent endpoint we control rather than to a per-token billing line.

Why this matters:

• The marginal cost of a meeting summary drops to roughly zero. A subscription is a flat monthly cost. Once it's paid, the hundredth summary costs the same as the first — nothing. A per-token product makes you ration; a flat-rate one makes you use it for everything.
• It composes with the rest of the fleet. The copilot isn't a standalone integration with its own keys and its own bill. It speaks to the same agent layer as everything else in the fleet, so a follow-up can become a CRM update or a ticket without leaving the system.
• No transcript leaves for a third-party meter. The audio and the transcript stay inside our own pipeline.

What it taught us

The build itself is modest — Swift, the system audio APIs, a transcript buffer, a call to our agent layer when the meeting ends. The lesson is the economic one: the unit cost of an AI feature determines whether people actually use it. A summary that costs two cents gets rationed. A summary that costs nothing-at-the-margin gets used on every call, which is the only way the feature is worth anything.

That's a pattern we keep reaching for in client work too — find the place where flat-rate beats per-token, and the product stops being something you ration and starts being something you lean on. It's the same insight behind the two-cent executive brief.