← Back

3 ingredients you need to build good agents.

Discovery, evals, and prompts.

Discovery: AI won’t replace us, but it will give us different jobs. Deeply understanding how humans use software is key to replicating those flows with agents. The UX fundamentals are still the same: clarity, empathy, and ease. I’ll admit it feels a bit strange to treat agents like humans.

Evals: a fancy term for “evaluation”, or testing. LLMs are non-deterministic, meaning that if you ask it “What is an eval?”, it’ll give you a slightly different answer every time. How do you ensure that you’re getting a useful, reliable answer that can be checked against your criteria? Evals.

Prompting: you already know how to prompt. Just type it in chat. But good prompting requires a lot more effort than saying “make it better.” In fact, when building agents, prompting is half the job. Even if you aren’t building agents, giving clear instructions is a core skill. Learn prompt engineering.