This AI is the new Crackberry

Hi, and happy Tuesday.

Remember the BlackBerry? Executives couldn't put it down. It was revolutionary: you could, for the first time, email from anywhere. 

And it was so addictive, people called it the “Crackberry”.

Claude Cowork is starting to show similarities to the “Crackberry” - but for all work.

Last week, Anthropic - the maker of Claude Cowork - added a "walkie-talkie" feature. You can be on your commute, at the gym, or at your kid's practice and either text or talk to Claude AI on your phone. Paired with Claude Cowork - running on your computer - it will then follow your instructions and operate your computer.

And, because your computer has access to your apps, your browser and your files, I’ve been able to ask it do do things such as:

• Open the Excel I was reviewing Friday and run that analysis

• Find that market report and answer three questions about it. 

• Start building a slide deck for me using this email

Claude - on your phone - will then share the results back to you.

_{A text message from your phone controlling your computer, as shown in Anthropic’s promotional video.}

This means we no longer need to be tethered to our desks to get work done.

Claude Cowork is doing for work what the BlackBerry did for email.

Microsoft has licensed this technology, so we can expect it to be available in Copilot before the end of the year.

(We’ll have to see how it’s priced and whether it will - like Copilot Chat is to ChatGPT - be a “mixed bag” equivalent using older models). 

But Cowork has a ceiling.

We've been building on OpenClaw - the open-source agent system that NVidia CEO Jensen Huang recently compared to Linux, the foundational operating system running 90%+ of internet servers.

Compared to OpenClaw, Claude CoWork has noticeable gaps:

(1) OpenClaw runs continuously. It doesn't wait for your instructions. You can give it a mission, and it can work in continuous bursts until it gets there. In contrast, Cowork follows your orders.

(2) OpenClaw includes a web server. It can build applications it thinks are necessary, and provide a web interface for you to use them. Cowork can't do that yet.

(3) And OpenClaw is open source. When we’ve hit limitations, we’ve been able to patch the code ourselves.

So while Cowork feels like a BlackBerry, OpenClaw feels like a colleague. 

But OpenClaw’s power comes at a cost.

(a) OpenClaw’s “self initiative” means most people set up a separate machine (or buy a Mac Mini) to use it, so it doesn’t wreck their computer. In contrast, Cowork is safer to use on your everyday computer.

(b) For the best results, OpenClaw needs to use Anthropic’s Opus 4.6 models. API fees for this can run $100+ per day. Cowork is bundled with Opus 4.6 access, maxing out at $200/month. (Our work-around is a complex mix-of-models setup).

(c) Cowork is easy to set up: you simply install an app on your phone and on your computer. Setting up OpenClaw is not as turnkey. 

This makes Claude Cowork the closest thing to a plug-and-play AI agent that exists today.

So the real question isn't which is better. It's which gap closes first.

Does Cowork grow up from a BlackBerry into a platform? 

Or does OpenClaw become more accessible?

We're betting it's both. We're building our “application layer” above OpenClaw to be portable. When Cowork, Microsoft Copilot, or whatever comes next reaches parity, we’ll be able to transfer what we build between them.

The first application we’ve built on OpenClaw is a monitoring agent that tracks topics tailored to highly specific interests - pulling from patents, industry news, journal articles, and more in real time. If you missed our recent demo, you can watch it here:

 AI as a Team Member - Demo

If you’re feeling adventurous, you can try Cowork this weekend. Install the mobile app. Give it a task from your couch. And let me know if you feel that Crackberry feeling… 

Best,

Dino