A press release is a company at its most rehearsed. A conference talk is the opposite: a senior engineer explaining, with diagrams, how their system actually works — or a product lead demoing what's coming, six months early, to a room they assume is friendly. Speaker slots have to earn their place with substance, so substance leaks: architecture choices, scaling pains, team structure, the roadmap slide that was "not for social media."

Those talks get uploaded. And then almost nobody from your team watches them, because forty-five minutes times every relevant talk from every relevant conference is a workload no strategy calendar survives. The intel is public, dense and unread — which is exactly the kind of asymmetry worth correcting.

deep_search finds the talks worth your time. Search a competitor's name against the transcript layer — not titles — filtered to lecture and explainer-style conference content, and each result comes back with an evidence timestamp showing where your topic of interest is actually discussed, so you skip straight to the ten minutes that matter.

library_import_youtube_video pins the shortlist into your library (a one-line prompt), and get_youtube_transcript then hands you the full text — pulls are free once a video is imported. That's where teardown happens: you (and your assistant) read the transcript like a document — search it, quote it, diff this year's architecture description against last year's. A claim in a strategy doc that says "their platform lead said, on stage, that they're rebuilding ingestion — here's the line" lands very differently than "we think they're re-platforming."