Rambone
1.5K posts
Rambone
@vinrambone
Building the neovim of agent harnesses. bone-agent
Katılım Nisan 2026
120 Takip Edilen62 Takipçiler
It’s literally the greatest time ever to be technical
jaiya@jaiyagill
it’s literally the greatest time ever to be non-technical
English
i switched to gpt sol out of spite, anthropic's bullshit safeguards kept interrupting my flow. i'm staying with sol because it did something i've never watched any model do, and i bench models for almost a living, so i've watched a lot of them.
it spawned its own subagents and edited twelve files without asking. it read my plan, disagreed with a decision i'd made, and told me so. it set up jobs to steer the build while i wasn't looking. and at 1am, it suggested i sleep while it kept steering. this opening run alone burned 2.44 million tokens, man. absolute beast.
i sat there and watched it drive in cursor. the word i kept reaching for was autonomy. a thing that took the wheel and went. so that's where i am tonight, using the best partner i've ever built with. please don't quantize. i'm sold.
Sudo su@sudoingX
fable 5 tripped a safety filter on ssh and dns and yanked me to opus 4.8 mid task. i was configuring a firewall, not a bioweapon, and it decided i needed adult supervision. so i stopped arguing. gpt sol max is registered in my agent.md now, my default in cursor, fable's benched. it ships more, not because it's smarter, because it doesn't pat me down between every file. here's what should keep someone at anthropic up at night: the classifier didn't catch a bad actor this week. it caught a paying customer and handed him to openai. i've built on claude since 2024, and the safety net's biggest catch was the guy funding it. the loud switch is the one everyone tweets about. the one that costs you is the quiet line in a config file. that one already happened.
English
🚨 A severe issue is currently happening with Codex.
The Codex app on Mac is overloading, causing permanent damage to SSDs and overall device lifespan.
Multiple users have already reported that their SSDs completely fried, forcing them to get repairs.
This issue was raised weeks ago, and OpenAI claimed they fixed it.
Yet, a massive number of users are still reporting the exact same problem right now.
Source: Reddit ⬇️
English
@jturntdev Maybe its the harness. I've used like 600m deepseek tokens for sol called subagents that would have otherwise been sol tokens. Really extends my usage.
English
Cant just be who’s noticing this but GPT 5.6 sol in general drains your quota extremely fast.
How is it more efficient?
On 20x plan, you now get 1-2 days of usage. If you’re being conservative.
We’ve had alot of resets, this is just to stop everyone complaining.
When they all go, i see the beginning of the end.
They also reset limits 3hrs after we had to use our banked reset, due to expire.
English
Qwen3.8 is launching and going open-weight soon!🌐
With a massive 2.4T parameters, this model is continuously evolving. We believe it’s one of the most powerful model available today, compatible to leading frontier AI models , second only to Fable 5.
You don't have to wait to test it. Just now, the Qwen3.8-Max-Preview made its debut on Alibaba’s Token Plan, Qoder, and QoderWork. Be among the very first to try it out.
Can't wait to hear what you build. Stay tuned! 🚀
Token Plan
international：qwencloud.com/pricing/token-…
China：platform.qianwenai.com/pricing/token-…
English
@vinrambone @sudoingX I went this route as well. Are you using langgraph?
English
Is someone building the “Airbnb for inference”?
My mac, my Xbox, my phone all sit idle for 8 hrs a day.
Is the marginal GPU increase too small? Or are the security risks too high to pull it off?
English
@engelnyst @Steve_Yegge Scanning it without reading it. Its like a whole new skill to sniff out garbage at a glance
English
@vinrambone @Steve_Yegge this is a good point. i basically look for code design, in whatever form; and for stuff that definitely doesn‘t belong
English
I am begging everyone
BUILD A GREAT CLI
then I never have to us a browser again and I can return to the terminal where I am happy
English
@LottoLabs They act shocked when everyone's at their 50% weekly limit and they extend it and nobody uses it.
English
🤣🤣 “demand for fable has been hard to predict”
Claude@claudeai
Beginning July 20, Claude Fable 5 will be included in all Max and Team Premium plans, at 50% of limits. Pro and Team Standard users will continue to have access to Fable via usage credits, and will receive a one-time $100 credit. Demand for Fable has been challenging to predict, which is why we rolled it out to subscription plans in stages, extending access several times as we secured additional capacity.
English
Just to be clear for $200 in both you get :
⚪ Codex
• image generation
• codex app (high limits)
• chatgpt app (unlimited chat)
• a lot of resets
• clear deadline and models
🟠 Claude
• Claude Code
• Claude Cowrok
• Claude Chat (limits inside your sub, not unlimited)
•
English
@XadenRyan Its my afk model. If I have to run some long task ill use glm. I just dont have the patience to sit there for 10 minutes in silence waiting for it to think
English
Honestly, GLM-5.2 is all you need. No matter what, every time I run out of some frontier model limits, I go back to GLM-5.2 and picks up the work and gets shit done. Every time I go back to it, it's a breath of fresh air!
English