What does Shadow actually do?
A lot of AI meeting tools still force the same brittle pattern: invite a bot, hope everyone tolerates it, then live with a transcript that misses whatever was actually being shown on screen. The same fragmentation shows up outside meetings too. You might dictate into one tool, summarize in another, and paste replies manually into a third. Shadow is going after that mess directly. Its core bet is that speech alone is not enough context, and raw transcription alone is not enough output. The valuable layer is seeing the meeting window, hearing the discussion, and turning both into the next action without making you reconstruct the scene afterward.
That is where the Skill model matters. Shadow does not stop at recording. It lets you define what should happen after a call or after a shortcut-triggered capture, whether that is meeting notes, a BANT breakdown, an email draft, cleaned-up voice typing, or another custom prompt. The product also claims useful supporting pieces around that flow: speaker identification, smart screenshots of meeting windows, markdown export, webhook delivery, and local-first transcription. Put together, that makes it feel less like a single-use note taker and more like a capture-to-output layer that can sit on top of everyday Mac work.