My AI agent was working… but I had no idea why it worked sometimes and completely fell apart othe...My AI agent was working… but I had no idea why it worked sometimes and completely fell apart othe...
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started
Back to feedPost
Taimoor's avatar
Taimoor Khan
• Dec 17
My AI agent was working… but I had no idea why it worked sometimes and completely fell apart other times. That’s when I plugged in LangSmith , and realized I’d been flying blind.
Suddenly, my “agentic AI” wasn’t a black box anymore, I could see everything:
Every tool call, LLM invocation and retry the agent made, also token burned & cached, state transition across agent steps
What surprised me most wasn’t the bugs , it was the waste. One trace showed my system prompt alone was 6.5k tokens. Not because it needed to be… but because I never saw it before.
So I fixed it.
Refactored prompts -> 1.5k tokens, same output quality Switched to dynamic system prompts, injected only when complexity demanded it
Lesson learned: If you’re building agentic systems without observability
AI Agent DevelopmentLangChainPrompt Engineering
Post image
Back to feed
The network for creativity
Join 1.25M professional creatives like you
Connect with clients, get discovered, and run your business 100% commission-free
Creatives on Contra have earned over $150M and we are just getting started