For the last year I’ve been following dozens of people on various socials and RSS feeds for artificial intelligence ideas and news. When someone shares a cool story using tools and techniques under the AI umbrella, I attempt to recreate it myself.
I want to see how they work. I want to personally explore and critique their potential value. I want to have a “library” of experiments to pull from as I encounter new obstacles. Experiments I can combine or extend from memory without needing time to figure out how to do them in the moment.
My best guess is that I’ve done this 50 times. Ranging from five minutes to an hour to recreate each one. Nothing has stuck. None of them have become a part of my daily routine.
I’ll give a quick example that feels typical. I’ll anonymize the people because that’s not the point. They’re awesome. The story is fun. The reaction is weird. And I fear for our sanity as an industry.
That last bit was sarcasm. I think… I hope…
Anyways, the example:
Their company’s website needed a wall of client logos to show off. They’ll have to go to each client’s website to look for their press area and hope they have an official, crisp logo to download. If that’s not available, their logo in the upper left corner, with any luck, could be a nice SVG. A few might be small and grainy JPGs, but there’s a chance they’ve promoted themselves elsewhere with a nice logo to steal. Tracking them all down could take time, but is doable. An email or two may need to be sent.
Getting 30 high-quality logos the old fashioned way might take an afternoon.
Or! We can ask one of the various chat tools what to do.
The directions given were roughly:
- Sign up for account with a logo database service (that I wasn’t aware existed)
- Hook Figma up to the service’s API (via Terminal with instructions included)
- Compile a list company names (in a text file or spreadsheet)
- Send text file or spreadsheet to logo database service (via Terminal with instructions included)
- Logos populate the connected Figma file!
Doing that took half an afternoon — roughly two hours. An experienced technical person could do that in few minutes, but I appreciate they wanted to learn how to do new things in new ways. They cut the time to do the work in half and gained valuable experience. Good for them!
Those last two steps. Wow! I bet that feels like magic. Hitting enter on a Terminal command and seeing logos start popping up in Figma. Hard not to feel like a wizard in that moment. Cooler than cool.
“I’m doing that right now!” I tell myself.
I go to sign up for the free account with the logo database service. While there I notice you can do a simple search for logos. I can download the logos. I don’t even need to sign up for the free account.
I search for the first ten logos that come to mind. I click download SVG for each. I go to the download folder in my computer. I drag my cursor to select them all. I press on the set of selected files and drag them into an empty Figma file. They scatter across Figma.
Three minutes. Tops.
What was the point of all that API and Terminal nonsense?
If I was regularly doing this with a much higher volume of logos, this automated process would a fantastic addition to my toolbox. Doing it once with a few logos? No. Laughably, hilariously, no.
I went back to the original story and saw comment after comment after comment congratulating them on their AI mastery and how they’re on the forefront of the design industry. A new unicorn AI designer is born!
Again! And I can’t stress this enough. The humans involved in this example are awesome and are learning like the rest of us.
But… My god… The environment around AI is incredibly dumb.
I remember this story more than others because it’s the one that made me realize 40 or more of the 50 experiments I’ve done over the last year were the same. A fun side-quest that made me feel like a wizard. In hindsight, there was always a better, easier, faster way to get the work done.
“AIs can find your syntax error 100x faster than you can.” — AI Is Like a Crappy Consultant
There are many, many, many activities certain technologies do faster than humans. Not marginally better but several multiples faster. Nearly everything I’ve bumped into in this realm of productivity I struggle to call artificial intelligence. They’re automation tools or services that can be done with existing technologies.
Like I was saying earlier, building these automation tools makes sense when in repetitive environments. If a software service needs to pull new logos at a moment’s notice in real time, hooking it up to a logo database’s API with the appropriate dimensions and colors applied is awesome.
It’s not AI. It’s an API. And it’s awesome. But it’s not AI.
I doubt many serious people would consider this example logo story a useful AI demonstration. Although, many, if not all, of the experiments I’ve done are presented as though they are AI. When results are written about publicly, many commenters are congratulating the AI skills involved.
I’ve been fooled a few times. Deep into the process of building an experiment — convinced I’m doing AI furreal this time — I realize it’s three simple automation tools in a trench coat accessed through a chat shell. Likely doable manually in less time if I knew the domain. Something I could have done in a few minutes if I realized logo databases existed!
Which is where I’ll leave this for now with the only positive note I could muster.
I like asking Claude (or whichever chat tool I haven’t tried in a while) for recommendations on how to do a broad, general activity with modern tools and techniques — I struggle to find relevant and useful tutorials with direct searches these days. I read the amalgamated recommendation and occasionally discover a new tool, tactic, or technique. Something I would have trouble describing if I wasn’t currently aware of it. If I didn’t know logo databases existed, I’m not sure I could have phrased a specific search term in a way that I would discover logo databases. Or, more likely, it would never have occurred to me to ask.
After all these experiments, it looks like something did stick. Now I regularly ask “what’s the coolest way to do [blah-bidy-blah]?” And sometimes I find a new non-AI way to do it.
That’s cool… I guess.
Coming soon?
I’m struggling to concretely articulate the consequences, both good and bad, of using more sophisticated tools like prototype builders, design system builders, and research assistants across teams of people. I’ll do that as soon as I wrestle these chaotic thoughts into submission. For now, the one word summary is: meh (with 1,000 caveats).
Related
There’s an excellent beginner’s programming book that can help you do all sorts of neat little automation tricks that you can buy or read online for free called AUTOMATE THE BORING STUFF WITH PYTHON.
≈