Three Months With Harry

I have been doing an experiment the past few months, and I can’t believe it’s already been three. I had signed up for the Claude Max plan, set up an Open Claw assistant, and tried out what could be done. It was set up as a Telegram account that everyone in my household could message.

Needless to say, I have been able to build stuff at a pace unlike before. Side effects include becoming constantly tremendously tired.

Fairly recently, Anthropic pulled the plug on letting people use their fixed price subscription on harnesses like Open Claw, so that’s been turned off for a while.

I’ve been trying to figure out what I have lost since. It’s kind of hard to gauge as we’ve been on Easter holidays, and a wave of sickness, so there hasn’t been much usage anyway.

I did manage to build one thing without Harry, our claw bot, and what I realized is that it’s the longer term context that is missing.

I’d built quite a few apps with my claw bot. I spent some serious time designing the first one upfront and building proper patterns with it. I had started using it with the aim to apply what I’d learned about event modelling and event sourcing. But that time actually paid off because the subsequent apps just became super easy to implement.

We would plan things using an emoji-based expression of event models that I could review on Telegram, then it would go ahead and implement it.

It was nice to be able to reuse functionality across apps. Most of the apps I created don’t require a sign up to use — they use admin/member PIN access instead — and it was fairly easy to say “use the same implementation as in project X” and it just built it.

Not having to reexplain things is such a time saver.

When they pulled the plug on the AI buffet, I tried using Claude Code purely, and for the most part it was okay if you are talking within a session.

It’s still pretty powerful, but I feel it’s a little stifled. I see a lot of people who continue to use Open Claw, and reading about all the things they are building to maximize its use is very impressive. I definitely am having some serious FOMO.

I’m going to try to look at getting back while not being fully dependent on online models.

Just reading Garry Tan’s gstack and gbrain is really exciting. He shares his setup openly because he thinks everyone should have access to this. Fully agree — I think people who have the ideas should definitely be enabled to maximize this tool. I also feel that they should have access to it in the first place.

I think there’s a huge imbalance going on between the different grades of AI adoption in the industry. It will be interesting to see where people end up in the next few months. I certainly hope that the tools will not become so expensive that only the privileged benefit from them.