Everyone that is upset about this should take note: you are not a (coding) customer at $20/mo. Their coding customers spend thousands per month (week!) on claude and it's growing faster than they can keep up with (source: I'm one of them, and I know many other like me. We're budgeting 10-20% of engineering salary spend on tokens). It sucks to no longer be able to code on the cheap anymore, but don't fool yourself into thinking you have any leverage here.
It's not the dictating, it's the clicking and the finding and addressing and the formalities. For a transactional email it's a lot, especially if there's substance to it like needing to click a link and make a choice about some options.
This is doing a bigger number on me than it has any right to.
...why are they all skewed, save for the buns that are already lopsided? Those I'll note are perfectly seated. Some are more skewed than others. Like the Big Mac is only slightly skewed.
Is there a pecking order to how skewed they are? Some social hierarchy of sandwiches?
Yes. I had a spare M1 Mini so I decided to set it up. YOLOed the entire thing and connected all the integrations, though I only ever use Opus/Sonnet. I have a dedicated Discord server I use to communicate with it.
It really is eye-opening how powerful it is once you connect your life to it. The biggest improvement is actually the trivial stuff — emailing contractors, accountants, etc. I no longer need to open another app, navigate six separate steps, and organize everything on the fly just to send one message. I stay in one place, organize my thoughts about what I’m trying to do, and the execution happens automatically.
For example, open claw will find the relevant threads, lookup the relevant details via web search, present them to me, give me a draft, I can review it, edit it, and send. 2 minutes instead of 20.
As a software engineer using AI daily, I think this is the real unlock — staying in a single context and not getting lost in the long tail of trivial details that fragment your attention before you ever get to the actual work.
The mental model shift is important too. It’s not that AI does all the work and we end up with agents emailing each other. It’s that AI automates the steps required to achieve your goals, so you can focus on the substance — like actually writing the email instead of navigating to the place where you write the email.
I also have GitHub and Vercel wired up, which means I can jot down an idea for a little productivity app for me and my family and it just appears a few minutes later, deployed and ready to use. That alone has been worth the setup.
AI didn't do the work, I did. Building up context is the part we actually have to put work into. I'm not saying it would be impossible, but boy would it be annoying to have to constantly reach a new assistant about your whole life.
Memory. I have built up so many scripts and crons and integrated little programs and memories with open claw it would be difficult to migrate to some other system.
The conversation around GPU lifecycles seems to be conflating the various shear rates within the data center. My layman understanding is that the old 3 year replacement cycle had more to do with some component, not necessarily the memory or the processor, going wrong for half of their units by 3 years, at which point GPUs were cheap enough and advancing faster enough that it was more cost effective to upgrade than to fix. However, that calculus changes completely when the GPU and the HBM are orders of magnitude more expensive than the rest of the system. I suspect that we will see repairs being done on on the various brittle bits of the system and the actual core expensive components will continue to operate much longer than 3 years.
To be clear, that is the advertised effects, not the side effects. The idea is that if you over eat you feel terrible, but also you won't want to overeat since you'll be full.
The only reason I want these things to be any smarter is because I need them to do more work over longer periods screwing up. The only reason I need them to do more work over long periods is because they are too slow to properly pair with.
If I could have it read more of my project in a single gulp and produce the 10-1000 lines of code I want in a few seconds, I wouldn't need it to go off and write the thousands of lines on its own in the background. But because even trivial changes can take minutes by the time it slurps up the right context and futzes with the linter and types, that ideal pair programmer loop is less attractive.