@fcatuhe@lumendriada +1 this is something I would benefit too. I also would love to be able to use the socket API for the local and remote instances running on the VPS as I am using the socket API to have an agent supervising other agents
@bcherny@DavidKPiano Personally, claude code feels much more of a blackbox now. It provides very little surface for extensibility over the harness. No thinking traces for Opus 4.7, no tool outputs anymore which makes it harder to understand and correct trajectories. It's fixable but should be OOTB
@jkeatn Strategies for scaling skills and Claude.md file in enterprise to prevent poorly designed skills, context pollution and conflicting instructions. Approach to security around skills & plugins artifacts
@softwaredoug What would be the form factor of such map for a massive codebase? Your post is timely I just watched some popular coding agents do around 100 tools calls just for exploring a codebase, quite often with 0 results and many retries.
I believe IDEs may not remain the primary interface for software development in the future.
Instead, we'll likely shift to new interfaces that enable developers to manage and orchestrate higher-level workflows with a focus on supervision and collobaration.
This week is a lot to digest for Software Engineers. We're entering new paradigmsโfrom conversational coding to delegating entire tasks to unsupervised agents that run in the background. We are also shifting from instructions based to tasks based interactions.
@sh_reya Can't attend the course (would have to) because of other events but would really appreciate if that is available somewhere in some form after it.
@juberti Hey Justin, is there any information available on the noise reduction and semantic VAD for the realtime API? also are the new transcribe models be available for transcription in the realtime API instead of whisper-1?
@alexalbert__ Is there a way to interact with an MCP server from a Claude artifact? I haven't dive into a full implementation yet but wondering if there's a sample or recipe that showcase this
This coming Tuesday, I will do something I haven't done in years: teach a class with @wandb.
We'll talk about
- Hardware for LLMs
- Building evals and "looking at your data" before picking models
- LLM inference servers
https://t.co/72iIZREt5z