What's this about?
A screenshot-and-hand-it-over interaction, launched with a list of intended uses: TL;DR this Slack thread, fill out this form, use this API reference to add a feature, pull the themes out of these X replies, find which DMs need a reply, draft follow-ups, turn notes into a deck, help me trim this video in iMovie, turn this recipe into a grocery list.
Why screenshots instead of integrations?
Because a screenshot works with everything. An integration needs an API, permission, a maintained connector and a business relationship — and it breaks when any of those change. A screenshot needs a keyboard shortcut. It is strictly worse in fidelity (no structure, no metadata, no ability to act) and strictly better in coverage, which is the same trade-off that makes GUI agents both general and fragile.
Why is it worth noticing?
- The example list is the actual product thesis: most of these are things that have an API nobody bothered to wire up.
- It is the consumer-facing sibling of the WebMCP push OpenAI is also running — one route makes sites agent-readable, the other gives up and reads the pixels.
- Everything in that list is also a small surface for the fabricated-evidence problem above: a screenshot is an unverifiable claim about what is on your screen.