Sabitlenmiş Tweet
Life is exceptionally hard when you actually try. This just means that you are hard too.
Success is just a constant battle to prove how Harder than life you can be.
English
Adrián Treviño
299 posts
Definitely I agree!
Btw Gary, I think one of the things I’ve noticed about the complexity of installing gbrain, is that the agent installing it, often doesn’t follow instructions to the end because it “thinks” that in the context of the user, not all functions are necessary
I think for the wide-adoption of gbrain, a initial /goal prompt for the user to run, would make it far easier to adopt and reduce the learning curve
In my case I love it and works amazing but often times I go back to the docs in the repo and realize maybe the agent didn’t implement something, and when asked, it answers it was either not implemented at all or implemented partially
I love gbrain and it’s worked amazing for me! Along with gstack I’ve shipped like never before. That’s why I wanted to surface this as it might help its development! And the possibility that it helps more people!
Amazing work
English
Metaprompting example I just did for GBrain to rewrite the README in the newest PR:
Read the README as a user who has never used GBrain and doesn't know anything about it. Does GBrain seem good? Do you know what it is? Does it make you feel powerful? Do you know how to get started?
Run an eval against this, and then rate it on 0 to 10, then give it a rating. Why isn't it a 10? Figure out what to do to make it a 10. Then implement it.
English
It’s one thing mainly. I feel like opus 4.7 is constantly seeking my approval far more than gpt5.5 and so it tends to overfit and/or over agree. Even when both have the same instruction to push and not agree with everything. Gpt5.5 tends to “feel” the freedom to disagree o push back
English
A big thing happening lately is that when rewinding with esc of clicking esc after sending a prompt in terminal no code flicker the text disappears while before the prompt either got canceled with a legend or got interrumpted or was re entered in the terminal to rewrite the prompt.
English
we're doing a lot more of this, hunting down some of the most annoying bugs in Claude Code
let me know if you have any white whales
ClaudeDevs@ClaudeDevs
In the last four Claude Code CLI releases, we’ve shipped 50+ stability and performance fixes. Faster resume, stable auth, lower memory, fewer hangs: 🧵
English
@garrytan That’s what we’re here for, thanks for putting gbrain out there! It’s been a massive change for the best! Will keep using it and work to make it better!
English
GBrain is still in experimental mode and not easy to use yet. It won’t be until it hits 1.0 which is not for a little bit!
Sorry for it not being ready for everyone yet. It turns out spiking out big projects is easy but it’s long polish process that you still need that is hard
English
This is what my OpenClaw does when I ask it questions. It’s amazing how good and useful it is once it has your context and you want to do map reduce type questions like this
I’m building all of this into GBrain so you can have my skills and code in your OpenClaw
English
I'm giving away the Claude Code skills we use to manage $300k/mo in ad spend at ColdIQ.
4X ROAS on $1M+ spent.
Ivan, our head of growth, built them off 300+ hours running ad campaigns for our clients. They run Google, Meta, and LinkedIn ads from the terminal in plain English:
→ bulk edits across platforms
→ custom audiences from CRM lists
→ creative fatigue detection before CTR dips
→ bid adjustments at scale
→ performance audits across periods
Reply "ads" and I'll send the full repo. Must be following.
English
Adrián Treviño retweetledi
Hi Garry! Would a seamless integration to Claude code help you? I use gbrain fully in Claude code, and I’m working on a pr, hope it serves you and all the community :)
Hope to send it today or tomorrow :) may need some adjustments from you to match your intent but hope it serves everyone as it did to me.
English
I am serious: I welcome a PR or even an issue. Help me be better.
And if you are building, please make JStack
Share your stuff. I want us all to be awesome.
jonah@jonahseguin
I deeply appreciate open source and engineering! The concerns I have with GStack are my opinion and not without merit. I don't doubt it can be helpful for some users. The sort of hype-cycle happening within AI where everyone on X hypes up these skills, loops, etc. without even actually trying them is what I have issues with. I use claude code daily for planning, engineering, and iterating. I will share my full thoughts in a separate post
English
Peter you literally inspired me to do it
Now we need to get everyone up to 100x to 500x speed
Peter Steinberger 🦞@steipete
@garrytan You’re shipping harder than I do these days!
English
@maminokotoo Te amo mi reina! Puede haber dicho más todavía!
Español
I previously ran a loop every 30 min. because I thought it costed too much. now with the 5 min cache, it actually costs me less to run it every 4 minutes since the reading cost is much less. instead of seeing th possitive, people only see the negative. adaptation is a skill too
Boris Cherny@bcherny
👋 1h prompt cache is nuanced actually. It costs more for cache writes, and less for cache reads. Whether you benefit from cheaper cache reads depends on your usage pattern -- context window size, whether the query is the main agent or subagent, etc. We have been testing a number of heuristics to give subscribers better prompt cache hit rates, which means lower token usage and lower latency, when it works. But this effect is far from uniform due to the nuance above. Say you use 1h cache for an agent, but only used the agent to make a single query -- in this case the 1h cache would be wasted and you'd be overcharged. At this point we have rolled out 1h prompt cache by default in a number of places for subscribers to optimize cache duration based on real usage patterns, but we actually keep it at 5m for many queries also (eg. subagents, which are rarely resumed so you'd be paying for them even though they do not benefit from 1h). We also are not defaulting API customers to 1h yet -- this needs more testing to make sure it's a net improvement on average. Separately, when we do this kind of experimentation, we use experiment gates that are cached client-side. When you turn off telemetry we also disable experiment gates -- we do not call home when telemetry is off -- so Claude reads the default value, which is 5m. We will soon be changing the client side default to 1h for a few queries, since we now feel good that it is a small token savings on average for those queries. We will also give you env vars to force 1h and 5m. In any case, the token savings is nowhere near 12x unfortunately. It is a small win though, that we have been in the process of rolling out to everyone. Hope the explanation helps. More here: #pricing" target="_blank" rel="nofollow noopener">platform.claude.com/docs/en/build-…
English
Apologies, the rant was necessary.
I wish entitlement wasn’t standard. It’s not healthy 4 growth
Looking back at cave man, I appreciate simply having an old fridge. At least I have one
If people looked back 3 years, even 1 year. They would appreciate having Claude code more.
English
If CC prints money for you like it does for me and YOU can’t wait 2 hrs for reset limits, pay extra and use the api. Or would you rather Max dies and we're stuck on Pro/API? And if you used weekly usage on day 3. For all it’s worth you already used minimum $4k worth of tokens
English
@claudeai Basically this is a perfect environment to setup agents that run on the cloud for clients. It would be unnecessary to have the already running-on-suscription agents here, but for clients where we can attribute the cost of the automation to the cost of fullfiling, these is golden
English
Introducing Claude Managed Agents: everything you need to build and deploy agents at scale.
It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days.
Now in public beta on the Claude Platform.
English
Basically this is a perfect environment to setup agents that run on the cloud for clients. It would be unnecessary to have the already running-on-suscription agents here, but for clients where we can attribute the cost of the automation to the cost of fullfiling, these is golden
Thariq@trq212
Managed Agents is the first 'agent in the cloud' API that has the right mix of simplicity and complexity. Implementation details like how you manage a sandbox are abstracted, but you have a lot of control over the actual execution of the model.
English
@trq212 Basically this is a perfect environment to setup agents that run on the cloud for clients. It would be unnecessary to have the already running-on-suscription agents here, but for clients where we can attribute the cost of the automation to the cost of fullfiling, these is golden
English
Managed Agents is the first 'agent in the cloud' API that has the right mix of simplicity and complexity.
Implementation details like how you manage a sandbox are abstracted, but you have a lot of control over the actual execution of the model.
Claude@claudeai
Introducing Claude Managed Agents: everything you need to build and deploy agents at scale. It pairs an agent harness tuned for performance with production infrastructure, so you can go from prototype to launch in days. Now in public beta on the Claude Platform.
English
@trq212 @KunchamSathwik The only case where I spun agents without clear evals is for concistency checks. I tend to do them with sonnet. Per Claude’s knowledge, both opus and sonnet are equally capable of discerning whether 1=1 or not across lines and documents haha
English
@trq212 @KunchamSathwik Normally what has served me is something as simple as telling Claude first to create evil criteria that’s when I don’t know what to measure specifically myself or there’s is something I suspect I don’t know exists. That way most of the time it has a clear verification goal.
English
done about 10 of these calls so far + looked at more transcripts
many learnings but one of the biggest is that it's very easy to spend a lot of tokens on open ended verification that doesn't make your output better
I'll try and write more on how to do it efficiently
Thariq@trq212
I want to do a few more of these calls. If your MAX 20x plan ran out of tokens unexpectedly early and you're willing to screenshare and run some prompts through Claude Code please comment. Trying to figure out how we can improve /usage to give more info.
English