What good looks like: the key to better work with AI

If you want to get the best out of AI, a well-crafted prompt isn’t enough. You need curiosity, a little patience, and above all, experience and taste.

Many people ask “How do I write a better prompt?” or “Which AI tool is the best?” but few address what really matters: how to sharpen your ability to recognise what a quality output looks like, and to produce work that is genuinely good.

Jason Droege, CEO of Scale AI, put it well. His company hires thousands of experts to train AI models by showing them what quality work looks like in their field:

“A large part of the improvement in every advanced AI model comes from AI companies hiring top experts who fill in the gaps in their knowledge and essentially show them what good looks like in every domain where people use them.”

Not long ago, evaluation meant being given two short stories and asked which was better. Today it takes hours. A web development expert shows the model what world-class code looks like. A leading physician walks it through the nuances of diagnosis.

This only proves that the value of working with AI lies in pairing it with deep domain knowledge, experience, and taste.

One of my colleagues and I have agreed on this point: taste doesn’t have to be innate. You build it through practice, repetition… in short, by doing your work properly and learning to do it as well as you possibly can.

I see this in my own work every day. Let me show you with two examples.

🔥 Example 1: Slide assistant

I wanted to build an assistant for creating presentation slides. I selected the best slides from my presentations, had AI analyse both the form (slide types: process, table, title) and the style (the tone and language I use), and turned that into a brief I iterated on more than ten times. Now I can feed any content into the AI and the assistant produces a presentation draft in exactly my style.

It works brilliantly, but only because I knew what a good presentation looked like. And because I invested the time to fine-tune the results.

🔥 Example 2: Interview assistant

I wanted to respond to interview questions faster while maintaining top quality. I uploaded previous interviews and transcripts of my talks into AI and built an extensive brief for my AI agent.

It generates answers, I review and refine them. It’s faster, and the outcome is better than if I started from scratch every time. But again, it only works because my know-how is well documented and I know what a good answer looks like.

In short: I’m the one who sets the bar. And I want AI to try to clear it.

Do you know what “good enough” looks like in your field? Here are a few tips to get there:

• Show AI quality: Give it your best work, benchmarks, and examples.
• Iterate: Don’t settle for the first output until it’s exactly right.
• Evaluate: You’re the one who says “this is it” or “this isn’t.”
• Keep learning: Reflect on outputs and improve deliberately.

AI is a tool. A world-class tool. But it needs you, your taste, your experience, and your ability to recognise quality.

Maybe it’s time to stop asking “how do I write a better prompt?” and start asking “how do I get better at what I do?” Because that will hold its value in any era.

FD