I've built a networking-focused feed called Network Nook that matched this post. Please take a look and I'd appreciate any feedback and

bsky.app/profile/did:...

I've built a networking-focused feed called Network Nook that matched this post. Please take a look and I'd appreciate any feedback and

bsky.app/profile/did:...

I've built a networking-focused feed called Network Nook that matched this post. Please take a look and I'd appreciate any feedback and

bsky.app/profile/did:...

I've built a networking-focused feed called Network Nook that matched this post. Please take a look and I'd appreciate any feedback and

bsky.app/profile/did:...

That doesn't match my mental model of Claude Code. I thought it was local with inference only in the cloud.

Yeah I'm confused. They look like separate implementations so I guess that means the Python one doesn't directly benefit from Claude Code hardening?

Thanks. I need to dig into this. I thought of Claude Code not as a tool to build upon but an app. Now I'm confused as to whether the two Claude agent SDKs (Python and Typescript) are completely separate implementations... Which seems unfortunate from a systems perspective.

And the Claude agent SDK?

Is this suggesting that products should be built using Claude instead of frameworks like PydanticAI or that internal chat agents should be?

As an example, there's a paper out there that describes how one can reduce disk traffic over NVMe by up to 90%+ by applying a 20 line patch to the linux kernel assuming one owns their own smart SSDs, can patch that firmware, and owns the whole stack up through the hypervisor

The perf gains are WILD