Who'll write the Pi plugins to support this?

Late to the party but Opus 5.5 is really good. Only wish for it was for it to be available by subscription in my beloved Pi. I would upgrade my plan to the big ‘ol one if it was. You listening @AnthropicAI ? 😅

GCP Cloud Run feels so abandoned. Huge missed opportunity

Great to see real engineering still happens

Started using GPT-6-Astra this week and for the first time I've run out on the $100 plan. Just spent my second banked reset 😅 Might have to go back to Sol. What are you guys doing?

I’m definitely the cynical type, but my guess is: LangDock’s primary customers are relatively conservative German companies and selling “AI” to them is much easier if you’re incorporated in Germany.

What I like the most about this is that by narrowing the scope, they made a 2 magnitude improvement to a - narrower than the general LLMs, but still huge - use case. That general pattern is worth taking note of. Similarly, @turbopuffer introduced per-namespace cache hydration from S3: search is fast for the “working set” of namespaces which is cached on NVMe but unlike the general solutions, you can’t search through all documents (across namespaces) quickly. This is perfect for a bunch of use cases and the solution makes it much cheaper for those situations.

I dont understand this flex. 1. Doesn’t talk about output. Would be like saying I burned 100 gallons of diesel today 2. Clearly unintended use of subscriptions. If enough post like this, they will have to clamp down on this.

Pretty sure @OpenAI always sits on it until right after @AnthropicAI releases their new model and then release it just so they will always have the newest model 🤔

Yes! And since agents are doing the driving, every user is now a power user. This means you should expose the whole surface through MCP/CLI

Been thinking about what the "Everything app/agent" will be. I think it's more and more likely to just end up being ChatGPT. The everything app is what you'll use for .. everything. Today, you have your local terminal based agent that you use for most coding tasks, and/or you might be building some stuff with agents in @linear based on customer feedback etc. and you probably use @ChatGPT to answer a bunch of professional and personal questions. But I'm increasingly convinced we'll end up just using a single app to talk to an agent which connects to everything else we do: code repos, email, calendar, banks, chat apps, support inbox: you name it. It has all the context. ChatGPT is not really there yet, but i think it's inevitable that something will become the central hub for everything and ChatGPT feels like the best positioned to do it?

Google Cloud Observability feels neglected or over engineered. Can’t imagine Googles internal tools are this convoluted? What’s going on here? It’s clearly below 80% but still sounded the alarm. I’m sure it’s my own fault somehow but not sure exactly how.