jb55's avatar
jb55 _@jb55.com 3 months ago
claude code + opus 4.5 instantly turned me from a vibe code skeptic to a believer. The planning mode + subagents are really good at exploring the codebase figuring out what to do. Its amazing what a little bit of context engineering can do. It just goes to show if you’re not staying on top of this stuff in realtime you will have a wildly outdated perspective of its usefulness.

Replies (63)

i've been using the cheap af chinese AI, z.ai - I paid $34 USD for one year I think, Black Friday sale. They kept it going for a while tho and probably will offer something similar for the Lunar New Year maybe? Not the Max plan but I imagine it will be 1/10th the price of Anthropic and the other American AI's.
Trying out clawd.bot currently. That thing is like claude code but on steroids. It's really impressive now that I'm starting to use it like a technical diary. It's giving me more feedback which I want to see and know about from my inquiries inbetween more token heavy questions Also Google's AntiGravity is quite INSANE. As an IDE and for planning it's amazing. It made a website for me, with placeholder imagery generated just for me, all on it's own without me really doing anything. A very simple prompt too no more than 2 sentences I think?
That was the one I watched today to that convinced me to spin my Damus Mac up to run it ;-) Although, I might see if it can run on Linux, assume it should be able to. I’d like to add TTS and create a virtual camera, and process audio feeds to text into it (have built that already) so it can join corporate Teams/Slack calls.
jb55's avatar
jb55 _@jb55.com 3 months ago
"it can control your computer and read your emails" uhhh i don't think i want it sending my emails to anthropic
True, but that's also a skill issue. I've been able to squeeze out a ton of code (100,000+ working lines) on a $10 plan but that's also because I've worked at tweaking how I prompt along with several different tools to assist. Now I'm imagining what I could do with max...🤩
its amazing. it can do increadible things even in the hands of a complete rookie. It would be amazing in the hands of a professional. Now will this get us faster #notedeck development? thats the real question :D
jb55's avatar
jb55 _@jb55.com 3 months ago
i mean i've been getting agents to do notedeck dev all weekend. i just need a way to control them remotely. so this might be useful for that. i was building my own thing in the meantime
I’ve literally never built a Mac app and Claude in Google Antigravity helped me cut 4 releases and a website for the app in a week 😳 Havenformac.btcforplebs.com The future is so bright!
We’ve been talking about this on Good stuff pod for months now. I have a similar personal setup to clawdbot I built at SovEng called @Wingman: Yolo Mode that I run on its own Mac Mini. Great it like a person. It’s own accounts, own npub so it can manage its own encryption etc, no access to my npub! We have other systems to delegate tasks from a Kanban into etc. I really should record a video on this setup, but it feels like it’s never static enough to document! https://fountain.fm/show/5c1jIyQLUHKkpS1r0ssb
Best part actually so far with clawd... is you can fuck off all reliance on providers and dickheads like anthropic (the ex nsa director is their coo or something i think?) or maybe that was chatgpt. Either way the consequances of using cloud based compute and AI is catastrophic for the development and builders who use it... and by extension anyone who has a interest in gaming on the pc or even engaging in free speech. The ram prices and gouging is going to hit so hard that the price of ram and vrams is probably going to go up indefinitely, on a exponential scale maybe and will outprice the consumer, placing it exclusively in the hands of moguls like Bezos who wants to have the planet rent their pc from him. Consequnces and vision im envisioning sadly is we will be pushed towards a cloud compute future, albeit slowly then suddenly. The restrictions on cloud compute will be minor at first, it will be unlimited usage and fairly priced. Then the ToS (?) will change overnight and there will be rations on the amount of hours you can rent. (GeForce Now / NVIDIA) has already started to enforce this policy which is very disturbing. "You will own nothing, and you will be happy." - perhaps our digital rights have been a target from the start too, along with our other rights. Oh shit, what a rant, my bad 😅😂. Ok so the tldr; use lmstudio and selfhost your AI. Invest in ram and vram like you invest in bitcoin. What use is bitcoin if your children and their children dont even have a pc to use bitcoin and freedom protocols on.
Local. Macbook Air but I recommend something with a chunky GPU if possible and a shitton of RAM... purely for the selfhosted options and to eacape bigtech and their spying. Im using GLM 4.7 and GLM 4.7-flash through and flash is self hosted. Z.ai was really cheap on black friday for 1 year.
i'm not really hahah... i've just thrown it on it's own subnet in my LAN and it talks to nothing but the nets and itself. I have little snitch going and I make sure it doesn't ping or talk to anything but the API servers and relays I approve. Other than that... not much really. Going to read the docs on security practices today I think now 😅