Nov 27 checkpoint

I feel like we've hit potentially the first "product-is-good-enough-for-x" point since we started. You can now Sign in to scout, sign in to your services, and ask Scout to do stuff within those services, and each of those steps works reasonably reliably.

Next Tasks

• Going methodically through the list of services like a checklist and documenting the extent to which they work & issues like screen sizing.

• For services that need desktop, don't have a 50/50 grid - instead switch to 75/25 split

Other Learnings

• Smooth isn't good at summarization tasks, it seems trained to be very utilitarian. For example if you ask it to summarize all the conversations, it will just give you back the raw transcripts, even if you prompt it to summarize. It will say things like "Action Items: ..." and then just repeat the raw messages. We have a few options to address this: 1. Parse the returned output with a different LLM. 2. Tell people to use hyperbrowser for summarization tasks, 3. Add a router ourselves which would check what kind of task it is and route to Hyperbrowser if summarization, or 4. Tell people for now not to use it for summarization.

• The "Skills first" approach feels less natural when I'm attempting to use the product each day. I find myself just selecting free text mode and then typing out the full url of the service. I'm wonder if it'd be better to simply allow people to @-mention services