Bryce Del Rio

Getting user feedback from our first class beta users: Claude - "Honestly, it's one of the better tool interfaces I've used. Here's my real feedback as an agent user: What I'd want as an agent: What surprised me (positively):" How long till agent reviews on websites?

@moguzbulbul @ebaqdesign @claudeai What are you all using it for? Has your workflow changed at all?

@ebaqdesign @claudeai Did the tokens in the $200 package also run out quickly?

Are you serious? Is anyone else having the same issue with their Claude limits? I ran out of my entire limit in just half an hour! I'm using a $100 package, and for the past two days, my limits have been depleting ridiculously quickly. @claudeai

I have this weird hunch to not sleep on the gemini cli.

@karpathy You're probably the best person to ask about this, is it because of the weight activation? Also I can't figure out what drives it's obsession, pigeonholing on it either way. But it's bizarre and seems like a relatively unspoken, unsolved problem with agent memory.

One common issue with personalization in all LLMs is how distracting memory seems to be for the models. A single question from 2 months ago about some topic can keep coming up as some kind of a deep interest of mine with undue mentions in perpetuity. Some kind of trying too hard.

We're here today in AI - "ok that makes sense, and how did it feel in filing? Smooth and congruent or were there questions along the way given the changes?" ● It was smooth. The one moment I paused was exactly what I flagged: The answer was obvious, but I noticed the question arising, which means the boundary is clear in my head now. Result - move clear boundary in head, to more free form for the future. In short, ask the model, and acclimate your docs to how it works naturally.

@sarahwooders @openclaw @opencode @conductor_build @greptile Simplest breakdown i think is this. We have a project that spawns task to headless claude for action. Fine. If I wanted to bring claude into that project directly in an agent loop and tried to bring my oauth token into the sdk and the agent loop instead of the api not ok.

I'm still so confused by Anthropic's policies for oauth tokens. The rule is 3rd parties cant use oauth tokens, so no @openclaw or @opencode. But for some reason, @conductor_build is OK So maybe it's that you have to use the Claude Code harness? Also no: @greptile can't use oauth tokens even though they use the Claude Agents SDK (which calls Claude Code under the hood) Very confusing and arbitrary

I breezed the article but a little confusing. Are they spawning agents from daemon? Or saying they've created an agentic daemon? Daemons a super powerful model for agents. Especially in the structure of handing off something to an agent who can do something with it or to trigger a web hook where an agent lives who can do something with it. But its conflating two things it seems unless im missing something. And its also missing a primitive layer, which is that instead of 100 daemons. You should just trigger daemon and let it pass to the right agent from a bench of agents.

I want to try this out. Looks interesting. There are some types of drift that I want to watch for. Mostly docs stagnating, but more than just that.

@bcherny @bytebrujo Everybody talks about dogfooding your use case into product. But very few talk about dogfooding your book into product.

@bytebrujo it me

Aint no way this is the same @bcherny that created claudecode.

I had Opus over constrained and we have first light of recursive self improvement. Have you ever wondered how Anthropic is shipping so fast? I have a glimpse that answers that and raises question if we have hit RSI. I had skills that were working decent but very verbose. Pointed Opus at it, said take best practices and rewrite them. It just finished a rewire and they are night and day. Long story short I noticed that Opus seems better at free form, or with capability to make decisions within certain guardrails rather than told exactly what to do. Our skills were very verbose in output basically line by line, from being burned in the past of not getting where we were steering the model. Now it seems that these are shortened guardrails in place, and claude seems like it enjoys them, and output is much stronger. I'll say again is this not the early stage of recursive self improvement. Also their docx skill is wild for other items.

@EthanChoi7 @OpenAI @AnthropicAI Isn't that just COGS for Anthropic after split?

The way @OpenAI and @AnthropicAI account for revenue / ARR is apples to oranges. Should Anthropic treat their revenue from AWS and other hyperscalers the same as OAI, they would be a materially lower in rev… If they both IPO in the coming quarters, not sure how the SEC is going to let these two companies have different accounting treatment for essentially the same type of revenue. OpenAI TAKES OUT the 80% revenue share that goes to @Microsoft Azure and others so reports this 3rd party revenue on a NET basis in their total revenue. Anthropic INCLUDES the revenue share that goes to @amazon AWS and others in their revenue so reports this 3rd party revenue on a GROSS basis in their total revenue. IMO, OpenAI taking more conservative approach that reflects the reality of the economics of these hyperscaler partnerships.

@morluto @MTabarrok Ooh nice

@MTabarrok worked around this by give it an OCR tool like GLM-vision one line + a ZAI key

still crazy how many tokens Claude Code uses just to figure out how to read pdfs every single time

@MTabarrok What i picture it doing when its writing up that sweet python on the fly.

PSA: you should have a personal assistant with claude and or other intelligence in the terminal.

@dani_avila7 Claude: "I know im doing this right"

GitHub is down. My agents are trying to push and have no idea how to handle it We need a hook that triggers whenever GitHub is down and tells Claude it can relax and go grab a coffee

@skylarromines @SJFriedl @DougCarlisle @cgtwts This

@SJFriedl @DougCarlisle @cgtwts If it’s true and there’s no solution to that problem then why keep creating something so detrimental to humanity

Anthropic CEO: “50% of all entry-level Lawyers, Consultants, and Finance Professionals will be completely wiped out within the next 1–5 years." grad students and junior hires are cooked.

@skylarromines @cgtwts Weird spot to be in for them. Imagine knowing it and not saying anything. Although they all seem to salivate at the idea, vs warning about it. Likely because the value of all human labor becomes an inference token with a meter they own.

@cgtwts Maybe this is not the best messaging to be putting out there. Imagine being a college student or in law school right now & hearing this. Extremely discouraging.

@altryne Wondering if this is from browser use dispatch or true bug. We haven't seen any meaningful change and things seem normal

My feed is showing me a bunch of folks who tapped out their whole usage limits on Mon/Tue. Is this your experience? Please comment, I want to understand how widespread this is

@trq212 The training feedback loop of this in iso is cool how it will drive phase 2.

turns out being an AI safety company is useful for when you need to make sure AIs can run safely

@bekacru Love this, were going to try this on an agentic platform weve been working on.

@Bouazizalex @howard @SlackHQ @Benioff @roam Claude migrate team make no mistakes

@howard @SlackHQ @Benioff @roam Full switch, 7k people?? 😅

We need @SlackHQ 's MCP to be so much better @Benioff 🙏