@jarodtaylor@itsolelehmann great work! I took your gist and the concept from the parent tweetm, added to it and created a similar Claude Code Skill https://t.co/RTZ4mIGoNf
I bet no one will use Cowork. Not because it isn't an incredible tool, but because people don't know how to use it.
Claude Code is the ultimate prosumer tool. The 1% of 1% who can effectively leverage it in every which way can accomplish so much more than their pre-AI selves. But normal users don't really know what to do with it. I wouldn't even say this is a divide between professionals with an engineering background and those without one because I've seen many non-technical users become very proficient with Claude Code. The divide to me is a deeper skill somewhere between proficiency with AI and simple curiosity.
There is a fundamentally different way to interact with a local agent versus a piece of traditional software. Feels like a push and pull thing.
Most software you input data and then take actions to pull that data out in various ways. Dashboards, charts, spreadsheets, whatever. The software holds data and you pull from it.
But agents you have to push. Push to find the right data, push to do the right thing, constantly nudge them along their way. This is a multi-step process that traditional software never had. Sure it can yield better results, but if you're bad at steering you never get there.
Thats why I find my family to not be so impressed with AI.
When I was home for Christmas I was talking about AI and in particular ChatGPT with my grandparents. My grandpa pulled out his phone, downloaded the app, and sent his first prompt. I forget what it was exactly but I instinctively stopped it and showed him how to improve it. The mistakes he made seemed so obvious to me but I think most people simply don't know how to prompt and use AI. This is the skill.
I couldn't even really explain what was wrong besides "you have to give it everything to solve the problem" and he said "isn't it supposed to do that?"
That one exchange kind of captures the whole problem.
Technical people like us assume the barrier is the terminal. We think normal people are intimidated by command lines and code-looking interfaces. So we build prettier wrappers thinking thats the fix. Cowork is exactly this, Claude Code with a nicer UI and a friendlier name.
But the real barrier was never the interface. It's the steering. It's knowing how to push. If you couldn't get value out of Claude Code, you probably won't get value out of Cowork either because the underlying interaction model is exactly the same.
The tool was never the problem. The skill is.
Interesting Claude Code:
"Scan my local network and find any interesting devices. Devise some interesting and useful things we could do with these devices. Check if they are hackable. By hackable, I mean we can ssh to them or connect to them via a shell or terminal. "
It’s amazing that no model has caught up with Claude on coding. Even if they look good on benchmark they’re still not as good at generating working good looking modern web apps.
Whatever magic Anthropic did seems very durable.
10 years ago today, I published my big post on AI. It has aged quite well if I may say so myself. Of the many 2015 predictions I cited in the post about the rate of AI progress ahead, it seems that the most bullish are the ones who had it right.
Honestly, at this point If you give me a programming interview and don't let me use AI assistance you won't get a very realistic idea of what I'm actually capable of