So the ‘AI Differentiator’ Is Taste … But What IS Taste?

How do we define “taste” as developers? And can we cultivate taste as a skill?

If your algorithm is anything like mine, then the following statement will sound (possibly painfully) familiar: “In the age of AI, when anyone can build anything, taste is the differentiator.”

It’s been echoed across think pieces, tweets, skeets and various forms of LinkedIn poetry for several months now, and perhaps rightly so. It certainly makes a compelling sound bite or headline (I’m adding my own think piece to the pile here, so there must be something to it). There’s a kind of existential fear in the air right now for software engineers (myself very much included): If an agent can code faster than me, better than me, cheaper than me or some combination thereof—what do I bring to the table?

There are, of course, many answers to this question: the historical context of past decisions made about a codebase, years of software building expertise and the learnings of past mistakes, the knowledge of when an LLM’s suggestion isn’t the right one, a deeper technical understanding of the problem, the recognition of when the rabbit hole has gone too deep and you need to try a new approach, the empathy you have for your users, your humanity, your craft, your love, the ability to step outside and feel the wind on your face, so on and so forth. Recently, this amalgamation of qualities has been distilled into one word: taste.

Taste is the ability to look at a thing and know if it is good. Taste is technically subjective, but we pretend that it’s not. For instance, my love of Mel Brooks and disinclination to watch any film over 90 minutes has led many people over the years to declare that I have bad taste in movies. Are they right? Possibly—although I like to think that I make up for it with my superior taste in video games (exclusively 80+ hour turn-based JRPGs). Point being, one person’s taste is another person’s distaste.

Taste is also temporary. The things we considered tasteful in 1999 are different from the things we consider tasteful today (the only known exception to this is translucent case electronics, a la the Apple iMac G3, which everyone agrees were great and should still be made today).

Still, there are clearly some qualities that rise above the noise enough to be considered in generally good taste. What does something have to have in order to become “good”? If the only thing separating our work from AI slop is taste, then our new primary task as developers needs to be figuring out what that means. What makes something tasteful? Is it even possible to define taste? (Spoiler alert: I think yes. Is that a hot take? Also maybe yes.)

Defining Taste

Often, when you ask someone what taste is, you get The Cop-Out Answer™️: “I’ll know it when I see it.” While that might be arguably true, it’s also patently unhelpful for anyone attempting to cultivate their own taste or create something tasteful. We may not be able to itemize every single aspect of taste, but I do believe that it’s possible to pin down at least a few. Here’s my attempt at capturing what I believe to be the five core aspects of tasteful work.

Attention to Detail

This one could potentially encompass all the others, since good work requires attention to detail as a prerequisite. However, I still think it’s worth calling out, especially since it’s one of the things that AI doesn’t do particularly well on its own. After all, what’s the first thing we’ve all been trained to do now when trying to identify whether a picture is AI or not? Zoom in and look at the details.

A website or application won’t have the same tells as an AI-generated piece of artwork (I hope), but the proof is usually still in the small, detailed work that AI simply doesn’t excel at. In an image, we might be looking at things like the text on signs in the background. In an interface, we’re looking at things like whether the same components are reused across different pages, whether the button text is still readable on hover when the background color changes, whether you can use keyboard navigation, etc.

A key aspect of taste is consistency. When we look at AI-generated websites, we often see a lot of inconsistency—especially in larger projects that include work generated by multiple different developers over a long stretch of time. Layouts and user patterns shift across different pages, icons have different sizes and line weights because they didn’t all come from a single icon set, the visual style doesn’t always exactly stick to the brand guide, etc.

We also see a lack of taste when users outside the assumed average aren’t taken into consideration. Can your users navigate with the keyboard, change the size of the text, undo actions they did by mistake or pause and restart various actions?

All of these are examples of small but extremely common user tasks that just happen to be outside the “happy path”—that default, assumed user flow where people do everything perfectly the first time (you know, the one that doesn’t actually exist). If you’re not paying attention to the details, these slip off the radar entirely because they are, by nature, not intrinsic to the core user flow. Instead, they’re what happens when users deviate from the flow, which is just as—if not more—important.

These are just a handful of examples. In reality, there are hundreds of tiny details across the entirety of a website or application that can make or break users’ experiences. Taste is not only the willingness to drill deep and pay attention to those small choices, but also the awareness that they exist in the first place and the knowledge of where to find them.

Identifies and Solves a Real Problem

Speaking of users: what actual problem is your software solving for them? What hole in the market does it fill? What can it do that no other product already does—or what can it do better than something else that already exists? Why would users reach for it and when would they need it? AI generation empowers the creation of software that can do anything, but more important than a tool that can hypothetically do anything is a tool that actually does something.

It’s ridiculously easy now to ask AI to make a copy of software that already exists (more on that in the next section), but that’s because when we prompt something like “make a copy of Minecraft,” we don’t have to be able to articulate the problem, the method, the goal, etc. We’re just telling AI to look at the problems other people have already solved and duplicate the solution.

Taste is the ability to recognize a problem, engineer a solution and describe the implementation of that solution to other people (and/or agents). If you can’t articulate what you’re solving for, then you’re just guessing—and AI is already great at that.

Excels at the Task

All that said, it’s not enough to understand and solve a problem. You also have to solve it well.

A recent AI clone of the Adobe suite has been making the news lately, as people (mostly people who get paid to write provocative headlines) wonder whether SaaS software is dead now that anyone can prompt their way to custom software that does whatever they want.

As a Millennial, I’m also an expert in killing various social norms and trends (chain restaurants, homeownership and the diamond industry, among others), at least according to the headlines of the past decade, which I believe gives me the authority to comment on this most recent “death.” So, here’s my $0.02:

Software clones have been around for a long time; AI might make them faster and cheaper to create, but that’s not enough to make people use them. If it was, GIMP or Photopea would have taken out Adobe years ago, and Photoshop would be relegated to the annals of design software history alongside Quark Xpress (pour one out).

So why haven’t they? Quite frankly, because they’re just not as good: they’re not as fully-featured, they don’t have all the same integrations and perks, their user experiences are more complicated and they’re buggy and inconsistent. Sure, they’re fine for some lower-lift use cases—I’ll reach for Photopea when I need to do quick file conversions or simple edits like removing backgrounds—but if I need to do heavy-lift mockups or big edits, the clones (AI-generated or not) just aren’t good enough … and the market agrees.

It’s easy to create something that does an OK job; it’s hard to create something that does an excellent job. The first output you get from an AI will probably do an OK job, but it’s the things you add on top of that output that push it into the realm of good taste.

Taste reflects the ability to empathize with our users and think about how they would actually use the software—not just how we hope the software will be used. This allows us to push beyond the first solution to the innovations that ultimately set the product apart.

Well Designed

Of course, we can’t talk about taste without mentioning visual design. The way something looks is probably the thing we most associate with the idea of taste—when work aligns with our preferences (or greater societal preferences), it becomes tasteful work. While visual design isn’t the only aspect of taste, it would be a lie to pretend it’s not part of the equation.

Good design supports the needs of the user while also adding an element of beauty to the work. Tasteful design can be a lot of different things—provocative, classic, minimalist, decorative, you name it—but the most important aspect is that it enables the user to complete tasks effectively. Once your design starts getting in the way of user work, it’s no longer in good taste; just ask Apple how people felt about their “Liquid Glass”/glassmorphism UI rollout last year.

The good news is that UI design is a learnable skill. If you’re interested in developing good design taste, there are more resources available than ever to educate yourself—including (brazen self-promo alert) my free e-book, Foundations of Design for Developers. Suffice it to say, gaining an understanding for visual hierarchy, color palettes, font selection, balance, etc. will put you at a distinct advantage when assessing work, whether that work is AI-generated or human created.

Taste is the ability to create visually beautiful work that doesn’t get in the way of the people using it.

The ‘Human’ Touch

Of all the things that make up “taste,” this one is admittedly the one with the most je ne sais quoi. What does it mean for something to feel human in the age of AI? Is that even possible? I would argue that it is, because it’s not necessarily about whether the finished product was AI-generated, but whether a human was involved enough in the ideation and creation to influence (and override, when necessary) the AI output.

We allow some level of imperfection in exchange for the knowledge that something was handcrafted. A hand-knit garment will look objectively less perfect than a factory-produced one, and that’s not only “all right” but often part of what we look for. The impression of the brush strokes on a painting are not seen as a detraction of the work as compared to a print; the shaping of a vessel that shows the imprint of the potter’s fingers is a design feature. Sometimes, the human touches that take us away from the statistical average are the things we seek out—the outliers that create interesting and unique work.

To be clear, I’m not arguing for adding artisanal bugs back into the code, just for the sake of proving that you wrote it by hand. Instead, I’m referring to the small touches that we can add in to showcase the personality of the creator: funny copy, easter eggs, moments of joy. The Chrome dinosaur game when the connection is interrupted. IMDb’s movie quotes 404 page. The flower explosion when you return a book early in Libby.

It’s not that AI can’t create these things, it’s that it won’t—unless specifically directed to, by a human. None of these things are intrinsic to the function of the software; none of them add a disruptive or product-defining feature. And yet, they could only be the work of a human creator, specifically because they are unnecessary.

Tasteful products are a reflection of the joy and creativity that goes into your work.

Cultivating Taste

If you’re tasteless, can you develop good taste? The answer, of course, is yes: you just have to be willing to work at it.

The good news is that we’re surrounded by interfaces and software literally all day. You probably already have some opinions about them, so now it’s just time to start thinking critically about those opinions.

• If there’s a piece of software you hate using, make a list of why. Write down every single task that you don’t enjoy and the friction points that make them difficult. Think about ways that you would improve it, if you were the maintainer.
• If you pull up a website and think, “Wow, that’s ugly,” take a screenshot, circle the parts you don’t care for, and explain what you don’t like about it. Was it too cluttered? Too bright or too dark? Too boring?
• When you read a line of copy and it makes you roll your eyes, think about what part of it didn’t resonate with you—did it sound cheesy, fake or patronizing? What words gave you that impression?

Start paying close attention to the systems that you work with, and noticing your reactions. In fact, go ahead and take a minute or two when you open a new website or app and just look at it—give yourself the time and space to have a reaction to it. Stay tuned into that mindset as you engage with it and think about how using it makes you feel.

That also includes paying attention to when things go well. Good products and experiences often disappear into the background (which is kind of the goal, honestly), so try keeping the same kinds of observation notes when you do work that feels easy, painless or even fun. For visual design, you could start keeping a scrap journal where you tape items that you like: interesting patterns, a nice combination of colors, etc. You can do the same thing on your phone with digital designs, just make a photo album or a Pinterest board and start collecting the things you like, like a little UI magpie.

Once you’ve gained that awareness, push a little further and start thinking about the why. Why do you think that designer or developer made that choice? What were the benefits and drawbacks of doing it this way? If you were to change one thing, what else would change as a ripple effect? How is this piece connected to the larger system? What are the maintenance requirements of this implementation, and would they be higher or lower if done a different way? When do I use this, and do I think other users use it in the same way? How is the experience different when I use this software on my phone vs. my laptop?

Taste is a muscle you build by regularly interacting with a thing, thinking critically about that thing and then verbalizing your conclusions (even if only to yourself—although it’s extra helpful to discuss with others, because they might have thought of things you hadn’t).

Having an understanding of the fundamentals and being able to put into words why you like what you like (and dislike what you dislike) is intrinsic to that process. Push yourself beyond “I just like it/don’t like it” or “It just felt hard/easy,” and see what you can come up with. Do this over and over again until it becomes second nature and is now just part of your experience using various software and systems. Make a habit of noticing and forming opinions, and start leveraging what you learn in the software you build.

Congratulations, you now have taste. Employ it with care; it’s evidently the only thing separating you from the clankers (for now).