Where should the AI CoE bot jump in?

My team is getting annihilated in the metric by this issue. It needs to be fixed. Just using some skin or worse app to do the work less efficiently as a workaround is a terrible idea. What does it take to get Claude code in various places appropriately recognized without sacrificing our ability to do the work?

@Amit Goyal and whoever gets to this... even today, that agent was not able to continue its work. What I did to fix was: Fork the conversation from the previous interactions If you try to fork from the last interaction, it doesn't work. It seems that last interaction somehow messed up with the OpenAI token. Meaning, it was not related to credits at all.

Getting a lot of 429 from Claude Code since yesterday, e.g. 429 {"type":"error","error":{"type":"rate_limit_error","message":"This request would exceed your organization's rate limit of 2,000,000 input tokens per minute (org: a3287c6f-33f5-43e5-8875-45c828f73510,      model: claude-sonnet-4-6). For details, refer to: https://docs.claude.com/en/api/rate-limits. You can see the response headers for current usage. Please reduce the prompt length or the maximum tokens          requested, or try again later. You may also contact sales at https://claude.com/contact-sales to discuss your options for a rate limit increase."},"request_id":"req_011Ca3FDMR5PvprgZLcn4sbL"}

I built a major new feature for Jarvis over the past few weeks and it continued to fail in production with real cases. Claude was down for a bit so I switched to Codex to patch what I thought was an easy fix just to get things moving again and a week later Codex is still finding random code, unwired code, duplicated paths, etc. It feels like Claude just "got it done" with no regards for my existing structure/architecture/strategy. He did whatever was easiest/fastest. Not what as best for long term code hygiene.

Are we getting new restrictions on Cursor? it's first time I see claude models not working

@Deniz Yavas or @Muhammad Usman : can you check on our Cursor spending settings? I'd be surprised if on demand spending is actually disabled, but that could be a quick fix

claude models on cursor having issues or is it anthropic in general?. use case I'm doing is building a full regression test builder autonomous agent

It's the usage caps. We hit the org rate limit, then restrictions were introduced. It shouldn't matter since it's all the same Anthropic LLM calls, but probably the Enterprise agreements are not structured with heavy development in mind.

I've been "manually" failing over to Portkey by removing the prefixed underscores whenever the Enterprise account hits the limit. (I just hit it btw.)

Our AI CoE - Go-To for AI Questions Office Hours workshop today covered: a) Switching some processes and workflows to Codex from Claude Code is easy b) Codex is arguably is a superior tool with arguably superior models c) Learning how to switch is a very valuable skill to obtain to always be able to stay on top of the best tools However, we ran into an issue that with our enterprise plan the usage limits are very low. Is there anything that can be done for that? I know that our AI spending is already high, but Codex is cheaper and maybe even better than Claude Code, so this could be an area for cost optimisation. But for it to save money, we have to have people try it first in parallel, so ironically what I am saying is we gotta spend money to save money

Do we have a strong evidence that Codex will end up both cheaper and better for our use-cases?

I have run into the rate limit every day with Codex - and it seems we can't do anything about it on our plan

I've been contemplating letting agents go loose and reducing micro-mgmt but it feels too risky, especially with production and admin-level access to many platforms. Any recommendations or good reads on how to fully delegate but keep critical systems safe?

Do we have any organization policy on using insecure agents like open claw agents that may easily leak confidential data to outside world?

Anyone else getting this in Claude Code? Alternates with Internal Server Error

Folks, is anyone else experiencing an issue where using the Terminal for Claude Code doesn't consistently count toward AI usage? I switched from Claude Desktop to the Terminal, and now my AI usage is taking a hit...🫠

How can we run out of credits if @Lucas Corbellini enabled auto-renewal? 🤔

same message came yesterday at the end of the day and today again, so even though it says forbidden, I believe it's because of some credits quota

Anybody else having issue on claude cowork? -

I'm on the Enterprise plan, but hitting the "You've used 75% of your usage" message in Claude. Can you clarify what the limit resets on, and whether our org admin can increase the allocation? I want to make sure I'm not cut off mid-work.

In one offs, it is less apparent. With a structured workflow, it is easier to notice the failure rates in certain ways. Things that worked reliably a month or so ago have started to show cracks without any explicit configuration changes*.

@Deniz Yavas - I have a team member who's stuck on a $100/mo token limit plan for Claude. She's raised a request 2 days ago but no response yet. Is there anything else she/I can do to expedite this?

Opus 4.7 is not an eager model; I have to coax it, which I never had to do with Opus 4.6.

I moved to claude enterprise account recently, I am getting this on claude code because I consumed the allocated usage. Who can help me with increase in the limit? Can provide explanation on the usage to the relevant team.

Sonnet is working fine, but Opus is having a glitch.

@Robert Egglestone I am using claude code inside Cursor. It is being tracked as cursor usage, can it be made to tracked as claude usage?

Do we differentiate between CC and Cursor usage? 😨

Trying to connect Google Drive to Claude... anyone know if this error is on the Google or Claude admin side?

It's a blocker for me, and though I have applied for API keys for OpenAI and Antrophic via Lumos, I'm still waiting for them to get approved. @Dmitry Degtyarev @Andrei Aiordachioaie @Chintan Parekh

hi team! I’m back to the AI credits problem… it seems I’m getting blocked on OpenAI too, not sure if because credits but likely yes. Anyone else facing similar issue? I’m on Warp using the OpenAI key and gpt-5.4 (high reasoning)model, this is the message: I'm sorry, I couldn't complete that request. Request failed with error: ErrorStatus(403, "<!doctype html><meta charset=\"utf-8\"><meta name=viewport content=\"width=device-width, initial-scale=1\"><title>403</title>403 Forbidden")

Seconded. Facing the same issue

The endpoint is being difficult and their documentation mostly just says it's in beta expect issues

@Andre Van Niekerk - could we extend this to cut our patent attorney legal costs and cycle time?

I feel like it is going to be a while before we can see a serious LLM running locally on a desktop/laptop. I have been experimenting with Gemma4 on an Apple M3 Max with 16 CPU Core, 40 GPU cores and 48GB of RAM. For the simplest of LLM asks, it has turned my machine into jet engine. If anyone has found a more efficient way to run a large LLM locally, please let me know. BTW - I was attempting the 31B model. Will be trying the e4b one next.

@Rahul Subramaniam can you show how much ram you are using while querying? If your LLM+prompt are pushing the total ram usage you might be doing memory swap and dumping your performance

Sonnet 4.6 has issues right now, swap to Opus

I'm on windows, using 2.5.20, and it's barely using 2.5 Gb or ram, maybe there are some corner cases where it spikes, or it depends on the repo causing it's indexing to go out of control or something 🤔

Hi Team, AI-First Win: 4-6 hours of human effort with 20hrs+ of agent run replaced 104* person-days of manual work. 816 test files. 90,894 lines of code. 14,468 tests at 99.4% pass rate. Zero test code written by hand. Zero build configuration built by humans. Coverage went from 0% verified to 50% measured.  Tests that previously could not run locally now complete in 18 minutes. Thanks to @Elena Loghin and @Bharat Kalra for driving this POC, and @Vlad Sorici for setting up the Gas Town initial knowledge base. Before Gas Town 460 test files existed in the repo but could not run locally — tests hung indefinitely No one could measure coverage because the test suite never completed locally Effectively 0% verified coverage After Gas Town 14,468 tests execute and complete in 18 minutes 50% i

Worth trying? https://www.thedeepview.com/articles/perplexity-may-have-built-a-better-openclaw

wow, when did the Perplexity max plan become so expensive!

and why are Grok and Kimi disabled?

Hi! need to create a Claude API key - who can grant me access?

Despite having premium access to Claude granted through Lumos, my account still appears as being on the Free plan.

there are two options one is for antropic and second one is claude