Hmm idk! I feel like excellence will always be valuable since it’s so rare. New era raises the bar.

But being good in two domains multiplies your value. That makes it hard to compete even if you’re excellent in one of those domains.

Also the typesetting and stuff is cool, I tried some book-related CSS a few years ago and there is a lot there. Even some dialects of CSS which are supported by certain ebook creation vendors IIRC

Really cool idea! Saw it on your table and wondered what it was

I’ve never used the Android app!

Hmm, I feel like I’ve been conditioned to disappointment with other models. Maybe I haven’t tried to put this one through its paces yet

I think I’m still most impressed by the layout engine. Best way to build interfaces by far

It’s like when you have two very different interests and you can combine them somehow, they multiply instead of add!

Shouldn’t be! I don’t remember the last time I was logged out

And then realizing you messed something up and having to do it all again 😓

Oooh I love that angle!

Another thing that’s great for progressive disclosure is models using --help on CLI tools. I feel like many MCP servers could be more like CLI tools.

This would be like wrapping a skill with an MCP rather than the other way around I guess

Great point! I suppose a skill only really requires one tool: run_<skill>_bash_command.

Skill metadata would be part of the tool description and instructions and resources could be retrieved by calling the tool.

I think the main shift is from “many specific tools” to “few general tools” (bash).

What do you think @tedunderwood.com?

Humans probably use some visual cues when glancing (headings, paragraph breaks etc), but mainly to find the most relevant snippets of text and then read those.

Interpreting a text as an image feels different and I’m not sure how to map it to human perception.

DO NOT RELY ON YOVR OWN KNOWLEDGE

It’s something I’ve wanted for a long time too!

Libraries like sqlx (or sqlx-ts) look pretty cool, though I haven’t used them yet.

github.com/launchbadge/...

Has anyone actually tried Agentforce? Is it a real thing that does something?

Do you allow the model to see its results and iterate in these tests? Would be really cool to see the difference!