Yep, observability is the key. Most solutions are not built for how agents can go wrong. One of the goals with breadcrumb.sh is to rethink how surfacing issues in agents should look

Still a lot of work to do, but you can try a demo of my new LLM tracing app here now breadcrumb.sh

You can now set up observations that run on incoming traces. These are essentially LLM as a judge evals, but much more flexible as you can have them monitor something specific described in natural language, or find annomalies on their own.

Writing docs is not the most exciting part of the process, but it means a Beta version is getting close breadcrumb.sh/docs/introduct…

btw here’s vercel’s ceo with a globally recognized genocidal war criminal Benjamin Netanyahu

What you are using for tracing your AI apps should be just as capable as what you are tracing. Completely redesigned the breadcrumb.sh trace detail page. Traces are now auto-analyzed for a quick overview and you can chat about them easily

producthunt.com/products/bread… Breadcrumb is live on Product Hunt 😸, go check it out!

Exploring an entirely new UX for breadcrumb.sh. Quite excited about this approach, will share more soon!

Progressing on this new direction for my LLM tracing tool. Agent tracing tools should be agentic themselves

Value is in automating insights, then go deeper. See x.com/i/status/20383…

pbr.studio New release incoming...

On its 50th birthday Apple is doing its best to become the most hated company in the world.

Introducing tryflare.dev Canvas mode! Got inspired by Paper to expand beyond basic CSS editing to a full infinite canvas with variant generation, comments and more! Simply launch it with the /flare-dev skill to directly push changes to your agent.