Jan-Stefan Janetzky

What granularity are your agents committing at? Atomic commits are a gift to future readers. Your agents pay for it now, in precious context. I cut mine down to one commit per deployable wave.

Bypassing the latest #EU #ageVerification app (2026.07-1) with a Chrome extension... again. Despite 3 months of security hardening and genuine improvements across the board, the fundamental issue cannot be solved. Anonymous age verification doesn't work.

@DeependraGaur15 Yes. Got nothing to hide regarding my Name.

Is your X username your ACTUAL name?

Hmm.. looking at hermes, where it's roughly 10x, I assume it has to do with compaction strategy. I use no auto compaction on claude at all. I simply optimized for very long sessions with fan-out sub-agents and loops. meanwhile, hermes compacts around 300k context using my own memory driver. I suspect this makes the biggest difference.

@iGottZ It makes no sense for it to be 50/50 since you input way more on each turn compared to the output, it makes perfect since for input to be 10x

According to cursors insights page, users use 10x input tokens compared to output tokens. My personal usage across codex and claude-code shows roughly 50/50 What about yours? (This is raw input and output. not cached, not thinking)

@elshayib_ tokscale.ai/u/GottZ

@MaximeRivest You mean that screen displaying this glorious lamp?

Hopefully nobody notices the 51 inches monitor behind 🤣🤷

You will think "RAG" but no.. I have spent a decade in document management. I have spent half a year to distil learnings from it into context management. I now run sessions in hermes with enabled compaction that do not forget even after 100 compactions. I've solved the context limit problem.

please i'm begging you show me something you built not another "this is my custom agent setup" post where you pretend you're doing something smarter than vanilla claude code please

@13_narcissus Thus, it teaches how to orchestrate models efficiently and successfully while giving enough domain knowledge to understand what's happening below the surface. Best of two worlds kind of.

@iGottZ Patterns is what AI excels at

Would you tell an 18-year-old (my son is 16 yo) to learn coding in 2026? yes or no — but defend it.

@TTrimoreau Either Hermes or Claude-Code.

What’s the first AI tool you open every morning?

@shinemeriz Apparently, if you mute accounts, it get's better too.

the worst part of the new algorithm update is that if you follow slop accounts your home timeline is about to get much worse

@DeependraGaur15 Provide a solution to a problem complex enough for someone to pay rather than build it themselves.

$0 budget. How do you get 100 paying users?

@sflorimm Sad truth but I also did not experience this amount of output ever before. It is a true accelerator.

haven’t met anyone working less after adopting the AI

@lumeroute @araseb_ go to meetups and talk to sponsors to get hired or referred.

@araseb_ Then what should I do?

1 hr of dsa 1 hr of system design 1 hr of applying on linkedin do this for a year and you'll still end up jobless.

@michellezfr And all because @theo told SF soybeans to talk to their models. Hopefully Theo has equity in these noise-cancelling masks. No for real tho.. Nobody could have seen this coming and Theo literally had a broken Hand.

Wearing noice cancelling masks to talk to Claude is crazy

@M5Stack Very nice!

The first batch of CardputerZero arrived from Factory

@ronsterd89 I'd take ridiculous selfies and put it back as if nothing happened

Ya spouse or significant other leaves their phone unlocked on the counter while their in shower what’s your next move ?!! BE HONEST

@elshayib_ that's how normalization with greedy whales looks like.

WTF Happned to SpaceX ? Since the IPO they lost over 1 trillion in market cap

@hiarun02 And yet, it requires you to get administrative approval to use teams on a quest.

Meta invested $73B so meetings could look like this.

@yagicchi86 WOW

あぁーー！！ できたぞ！ この透明度！！ これ研磨したらどうなるんや✨