Hey there,

Time for another weekly recap!

My AI usage has largely been shifting to Grok Bot. I've been setting up a bunch more systems and processes to automate my business there.

The team has also shipped a bunch of updates, and it's becoming a really useful and powerful tool for me.

I still use the other apps, of course. ChatGPT for image generation and browser use, because I think its built-in browser is still the best. And I still use Claude's Opus model for editing videos and for coding, either through Claude Code or Cursor.

But I think more people should try Grok Bot.

This week, Anthropic released Claude Duet, their own version of an always-on agent. The competition is getting tough. Now every major lab other than Google has its own version.

And it does seem like this is the future of interacting with AI. Creating chats and threads is starting to feel like a thing of the past.

If you want to see exactly how I'm using Grok Bot, including my tips and tricks, I put it all in my latest YouTube video:

I've been thinking more about product lock-in recently, and how easy it's been to switch between AI products.

From the AI labs' perspective, I think this has always been one of their challenges.

Sam Altman actually talked about this in his interview with Garry Tan at Y Combinator. He said that in the early days of ChatGPT, it was kind of clear they wouldn't have any of the traditional moats. It was just too easy to switch to other providers. The product wasn't sticky and didn't give people much reason to stay, but they kind of ignored it and kept going.

Well, I think some stickiness is starting to emerge now with always-on agents.

When Claude Duet came out, my immediate reaction was: "Oh, I don't want to set up my entire Grok Bot team all over again. That would be so much work."

If you think about it, these are the things I'd have to migrate:

Now granted, some of these might not be too hard to migrate. Your agents could mostly take care of them.

But I do think connections are pretty hard to migrate, unless you're using something like Composio, which aggregates your connections. Routines aren't super easy to move either because they work differently in different agents, and I think exporting and importing memory will always be lossy.

I also wonder if the AI labs will deliberately make things harder to export in the future. Memory, for example, which Grok Bot is already doing a bit just through the lack of transparency into the memory system. Or maybe they'll disallow tools like Composio that make it easier to migrate.

Because competing just on models and features is going to be a very hard game to play.

Btw I'm building up a Grok Bot course inside the community. I’ll be adding more lessons to it as we go. If you want to come help shape it, join us in the community here.

I was lucky to get invited by friend to an Nvidia creator dinner this week.

I met several people I already knew through Twitter including Kim, which was amazing.

And the goody bags they gave out are just next level 😄 I didn’t expect this.

I'm the kind of person who really likes just working and being with my computer, and not necessarily socializing.

But this event kind of taught me how important it is to do this and hear what others in the field are experiencing. I'll definitely be doing it more often.

Have a great weekend everyone!

Moritz