Andy Halliday

Pulse is the open AI assistant that's available to the high level subscriptions, right?

It's not available to a regular plus user.

A candidate for an emerging sort of independently proactive agent is Anthropic Cowork.

Yes.

And today they just announced that, or maybe yesterday, but they expanded the Cowork platform by adding agentic plugins.

And that lets non-technical users...

automate real work using it i mean when we say automate real work we mean okay let the agent take actions and automate things that you might otherwise have to you know take action and direct that you know an assistant to do so they have plugins now cloud code has always had a plugins directory and you can use plugins in cloud code but now there's a nice integration of

the concept of plugins and skills inside cowork, and they have already packaged up for you special plugins that basically, you know, make the Claude system, you know, a specialized knowledge worker in that particular field.

Yeah.

So you can add the plug in for marketing, for legal, for sales, for finance, for data analysis, for customer support, for product management and biology research.

So those are all available and you can make your Claude coworker skillful in those areas.

And then on top of that, you can add skills.

Yeah, let me weave off of superintelligence there.

There's more to be said about superintelligence, and we'll save that for tomorrow because there's a number of sort of people, high-level researchers that are leaving the main frontier model developers and going out to develop superintelligence.

And that's kind of the next frontier for AI.

But I wanted to talk about Malt Book and how it relates to superintelligence.

So in answer to the question, what is this for?

What's the purpose of it?

Well, it was put out there without any express purpose.

But you can see how it's a framework for collaboration among multiple agents, which may be the context in which superintelligence emerges, not from one single model or one single agent, but rather from the activities of multiple agents who are all prompted correctly