AI agent performance tracking

[Eggplant]

Our dev squad is stoked about bringing our new AI agent into our daily workflow. It sounds great on paper, but getting it to actually do the right thing every time is a ma###ive project. We're constantly throwing different prompts at it to see if it levels up or just starts spitting out total nonsense. Everything needs to be polished and perfect before we let it loose on real tasks. What's a good tool to track all these interactions and figure out if the agent is actually getting better?