
Coffee, AI and Taste
Are you a coffee snob? I'm a coffee snob, and honestly, I'm here to tell you that being a coffee snob is fun.
And look, I know the word "snob" carries a pretty negative connotation. It conjures up someone sighing dramatically about grind size or judging you for putting creamer in your cup. That's not really what I mean. (Okay, mostly not what I mean.) What I'm talking about is knowing what genuinely great coffee tastes like, noticing why a cup works or doesn't, and getting that little spark of joy when someone nails it.
Once you have that, coffee stops being just a way to get caffeine. It gets interesting.
You have to drink bad cups too
Look, no one starts their coffee journey able to pick out the citrus notes in an Ethiopian Gesha (that's some first-rate coffee snobbery right there). And you definitely don't get there by drinking a perfectly adequate cup every day. You've got to drink really great coffee to develop a palate for it.
But you also build taste with contrast.
You drink the burned pond sludge that some places serve (I won't name names 😉). You have the weird, sour cup that's pretty awful. You also drink the perfectly fine drip coffee at your local coffee shop. And then someone serves you a cup that stops you in your tracks and makes you say, "I had no idea coffee could taste like that! I can drink this!"
Both ends of the continuum matter. The great cups show you how good coffee can be and raise your expectations. The bad cups sharpen your senses and help you spot what went wrong and how the cup could be better. The extremes of the scale build your taste. The stuff in the middle really doesn't.
What does this have to do with AI?
What's true for coffee is true for pretty much all the stuff humans make. Books, music, software, writing, movies… you name it. For all of these things, you learn to tell the really good from the really bad by consuming both.
AI is now making so much of what humans have been making, and it does a perfectly competent job at a lot of it. It's the proverbial 7/10. Sometimes it's a 9. Sometimes it's a 2. Mostly, it sits somewhere in the middle. It's pretty amazing that you can generate so much reasonably good stuff on demand. It also creates a real risk of losing taste or never developing it in the first place. If we let AI become our main source of content, we're going to spend a lot of time in the middle of the quality continuum. We'll forget what it's like to hear music that's so bad you have a visceral reaction to it. We'll forget what it's like to read a passage so simple and beautiful that it stirs something deep inside you. We'll lose our sense of what's great and what's bad.
We'll lose our palate.
Working with AI
If you're following the current advice on getting good at using AI, you'll know that losing your palate (your taste) is the last thing you want to do.
It's easy to spend all our time learning about the latest models and the cool new harnesses and applications we can use with them. I spend a lot of my days looking at techniques and ideas to get better at the technical side of working with AI. But as models get better at producing work that's good enough, our most important skill becomes taste.
Whether you're collaborating with AI or working in "conjurer mode," so much of the work is generating ideas and content and then evaluating what the agent produces. You ask, "Is this any good? Is this mediocre wearing good's clothes? How can this be better?" Answering those questions well requires taste. You can't push a model toward a 9 or a 10 if you don't know what a 9 or a 10 looks like.
Consume without the model
One of the best things you can do to get better at using AI is to go without it sometimes. What a paradox!
It's really tempting to optimize for speed and efficiency by having an LLM summarize a book for you, or by pasting in a link to a long-form article and asking for three bullet points. It feels like a pro life hack to take the transcript of an hour-long podcast and have the model give you the CliffsNotes. These shortcuts may save you a little time in the near term, but they cost you your taste in the long run.
Read the book cover to cover. Listen to the whole album. Finish the entire blog post. Watch the video all the way to the end. Even if the content is weird, disconnected, and a bit clunky, you're developing your taste. Notice what makes things great and what makes them feel off. Spend some time thinking about what you just read, heard, or watched before you have the model summarize it for you. Make space for taste to develop.
So be a coffee snob, or a book snob, or a movie snob. It's fun AND it turns out it makes you better at using AI.
Related Posts
My AI Toolkit (Dec 2025)
The AI tools I’m using in my day-to-day as of December 2025. It’s been a wild year in terms of available tools and their levels of efficacy and sophistication. Now seems like a good time to reflect on my current usage.
Balancing Thinking with Agentic Coding
One of the most exciting things about agentic coding is the ability to move quickly. One of the biggest dangers of agentic coding is the ability to move quickly. One of the most important problems to solve right now is finding the balance between speed and learning.
Maintaining Excellence as a Software Engineer
In engineering, everything seems to have changed and we aren't going back. Nor do I think we should. But in the midst of all the change, and all the speed and all the hype, something quite important has changed and if engineers aren't careful, we might just find ourselves sacrificing our future careers (and minds) to go faster today