How to enable full observability and automatic analytics for your LLM-based application. It takes one library + one line of code, and you get a ton of information for free. This is a no-brainer.

This is likely the most impactful update you can make to your agent. Most companies that hire me have no idea how their customers are using their agents. More importantly, they don't know whether agents are delivering on their promises. They have zero visibility and analytics about their agents. This is always my first recommendation and the one that makes everyone's life easier. Voker is a very simple solution. They are partnering with me on this post. What I like about them is their simplicity: • Simple to install (one library and one line of code) • Simple to operate (a simple dashboard for the entire team) • Built for conversational agent products In return, you get automatic analytics with insights about how your app is running and where it is failing. Building this from scratch would take weeks. Here, you get it all without moving a finger. You can check them out here: Their free tier offers unlimited seats and lets you log 2,000 events every month, so you have plenty to experiment with.

I like this. The failure mode with agents usually isn’t “a user clicked the wrong button.” It’s the agent took six reasonable-looking steps, picked the wrong tool on step four, retried it twice, and now nobody knows whether the issue was the prompt, the tool, the data, or the app.

Santiago always give us alpha research as usual

ngl vendor-sponsored "no-brainer" is still a vendor pitch

Speculative decoding is one of those rare optimizations where the engineering complexity can pay off quickly, especially when latency matters more than peak throughput.

？Idk…… What happens？

This is what we needed rather than finding error in long langfuse tracing

This could really simplify monitoring for LLM apps!