Your AI shoehorn is the wrong shape

This might get a little ranty.

You can’t log in to any product these days without AI buttons, magic icons (there’s a whole other article to be written about how and why the ‘magic wand’ became the symbol for AI), and email newsletters declaring that “our product now comes with ADDED AI!”.

It feels like we’re reaching a saturation point of magical write shit for me buttons. Generative AI is easy enough to integrate into any box where a human would normally write something, thus, it’s becoming ubiquitous.

The problem is, we’re not actually changing anything. We’re just augmenting behaviours we already exhibit with something that makes it automatic.

Which we then have to re-read, correct, and probably rewrite anyway.

Products are doing AI wrong

Most of the more prominent implementations of AI are using generative AI, which takes some text as an input, and rewrites it, answers it, or adapts it in some way. There are some interesting uses, and having it available near your inputs seems to make some sense.

Miro, for example, now incorporates a Miro Assist button into their context menu so when you happen to right-click a sticky note, you might just spot the starry magic button and decide to make your sticky note more business for some reason.

I’m picking on Miro a bit here, but there are plenty of other apps doing this and I barely need to mention it, because you’ll have already seen - and definitely not used - them.

Recently I tried to use Retool’s AI code writer and it is just terrible. It takes a prompt and tries to write code for you. But it doesn’t seem to understand any context so it just blurts out a load of nonsense that never seems to work.

For AI to be useful, someone using the platform needs to know that these options are available and, so they’re not shooting in the dark, that the output will be useful. Otherwise, using AI feels a bit like giving your writing to a copywriter you’ve never spoken to, without context and understanding, and they could happen to get the output absolutely bang on or might just serve up an absolute pile of robotic nonsense.

The best results in any interface is the result we expect. That’s why, when we conduct usability testing and someone does something unexpected, we ask them: “What did you expect to happen there?”

Humans make shortcuts in our brains all the time, and when things don’t happen as they expect, especially when we’re trying to be productive, we lose trust in the systems we use and drop out altogether.

If you’ve ever described something as intuitive then you know what I mean.

The best AI is invisible

Thing is, AI has been around for ages. You just didn’t notice, because we didn’t call it that. If we apply the term to historical technologies, it’s everywhere - because it’s not really ‘artificial intelligence’, it’s just ‘crunching a buttload of data together and taking actions on it automatically’.

So really, today’s generative AI isn’t really artificial intelligence either. It’s a very clever, very fast version of autosuggest.

But there are loads of other examples of very similar technologies, which are designed so they work seamlessly in their context so they just make your life easier or give you a nice feature that just works.

On Android and iOS apps for example, you can cut out the subject of a photo by holding your finger down on them. Seamless, simple, and the design is sweet. Technically speaking, it’s using AI to detect the subject, cut out the background intelligently, and serve up a new transparent image. It just does it without having to be advertised in an ‘AI IMAGE MANIPULATION MENU’.

AI monitors your bank transactions and will stop things that look fraudulent. AI lets iPhone owners unlock their phones using their face. On a much smaller scale (in most cases), AI recommends products to you based on things you bought in the past.

As Amazon can attest, some websites do that much better than others.

The thing is, people don’t really want or need AI. AI itself isn’t a feature. It’s a tool that we can use to make lives easier, much like anything in tech.

Bad design will incorporate AI as a feature, which makes it hard and confusing to use, and unlikely to reach adoption on any large scale.

The best designs, by contrast, will incorporate AI silently and smartly into interfaces - just like the squiggly line underneath Grammarly-detected misspellings, or an autosuggested string of words after you start a sentence in Google Docs.

How to shoehorn AI into your product (properly)

If you’re thinking about incorporating AI into your product, here’s a quick step-by-step guide on how to go about it and what to think about:

1. Stop

2. Fix your usability issues first
   You should be doing this anyway, but test your product with users. Work out what works and what doesn’t. Know them through and through. Work out if there are things they’re doing that could be solved with automation

3. Understand the capabilities and limitations of AI
   Maybe do a course, talk to an AI expert (Chris Ballard of Justified AI is very good), or hop on to a webinar

4. Get a good product designer to help
   If you’re even considering building AI into your platform, a good designer will help you implement it well using contextual research, behavioural understanding, and good journey design (I know a good product designer too. It’s me 😉)

5. Prototype, test, and iterate

   It’s super fast to build prototypes these days and lots of platforms that can help, like Retool, FlutterFlow or Bubble

Just remember to treat AI as it is. It’s a tool, just like a database or coding language or API.

Use it when it’ll solve a problem. Leave it out when it won’t.