Generative AI today makes me think it’s 1997 all over again

Websites were a dog’s breakfast back then, but they had to be to get here now

Update March 26: Long-time colleague Richard MacManus wrote on Cybercultural about the internet in 1995. It’s worth a read to set the stage for this post. Oh frames, how we loved and hated you so.

I started building websites around 1994-95. I don’t remember exactly when, but I do remember writing HTML by hand in SimpleText, BBEdit, and other text editors. Visual editors were just coming out. Microsoft FrontPage was not that great and was infamous for creating bloated code. Everything was bespoke. Most of the things that are easy with CSS were done with (buggy) Javascript, set with rudimentary style controls, or just impossible.

The first WYSIWYG editor I remember using was Claris Home Page and after that GoLive Cyberstudio (later scooped up by Adobe and named just GoLive). They were amazing. I could build, test (ish), and create websites in a fraction of the time. I remember automating whole sections of websites with “templates” (nothing like today’s templates that any CMS uses) and being able to make mass updates without mass find and replace.

Oh those were the days.

But, as WYSIWYG tools hit the mass market, websites kinda went to crap. Just like all technologies, the web started out with a few folks who had arcane knowledge of how to make websites work. We didn’t just code websites, we usually had to set up the servers too. Knowing an Apache config file was pretty important back then. Or in my case a Primehost (AOL’s hosting company) config which I blithely decided to edit live on the server without a backup (oh youth). WYSIWYG tools were awesome. They let people who were actual designers help design websites. But democratized tools landed us into the world of “oh my nephew can build that for us for cheap” and it was a bumpy ride for the next five years.

Back then I was doing some web dev on the side, and I couldn’t compete with the nephew in college making websites for beer money. When I started at Glaxo Wellcome, the U.S. website was literally hosted on a computer in some guy’s garage in Denver. People got what they paid for. Professionals got pushed out and a lot of really bad websites were built. Eventually companies figured out that “you get what you pay for” didn’t do much for their brand and came back to agencies and professional shops. It wasn’t until 2008 that WordPress started make things easier for people and another ten years before people could make a decent website themselves without needing to know CSS or any of the arcane knowledge of the early 2000s.

I think generative AI is right now firmly in the era of websites in 1997. It’s everywhere. Some is great. Some is crap. But at least their websites look good.

This isn’t a bad thing

I’m not sitting here as an AI curmudgeon. I love using generative AI. The featured image for this post was done with Microsoft Designer (the irony is not lost on me). I’ll probably have Apple Intelligence take a quick boo at this post before I publish. I used ChatGPT to convert my resume into an ATS-friendly format. LinkedIn and ChatGPT give me a good start on cover letters for job applications (I am available for freelance work if you’re wondering). I ask AI to do a lot of things for me and I’ll certainly be one of those people who keep their jobs exactly because they can use AI effectively (current fun-employment excepted—let’s not go theres shall we).

But I’m also seeing generative AI everywhere. Not just in tools like Notion and Asana, but in the website I’m building with Circles of AI. “Here just let us create your whole website with AI. It’s easy! And you’ll have a great site in a few minutes!”

And this is going to make the ugly, poorly coded websites of the early 2000s look pretty quaint. And safer.

And it’s 1997 all over again.

The next 18 months are going to be a mess for generative AI. There are going to be some amazing tools that will make our jaws drop—heck there are a bunch of those already. There are also going to be AI agents that go rogue or are compromised by malicious agents. Well meaning people are going to use DIY toolkits to make simple agents for themselves that are going to take down companies and lead to massive data breaches—beyond the harm they are going to do to themselves.

It’s not going to be fun for a lot of people or companies.

What we’ll then see is a “collapse” in the AI space. Well, at least that’s how the press will cover it. It won’t actually be a “collapse,” what it will really be is the snap back to folks whose jobs it is to do this work. The AI equivalents of FrontPage will quietly go away and we’ll all go back to the drawing board to create the next generation of DIY AI tools that will make what we have today look like cute toys.

We’ll still see “let AI help you with your website copy” tools (and I think this is a good thing, btw), but they will be toned down. It will be less “let us take care of this pesky job for you” and more “let’s build this together…” I don’t know if we’ll see AI integration roll back from apps, it’s too soon to know that for sure, but if AI is in Notion or Asana or Mail or whatever, it’s going to be more subtle.

And it’s going to be more powerful.

The world of Her is getting closer, even if it’s just digesting all the email I got overnight and telling me what’s important.

This is the exciting time to be working in AI

Just I like I can’t code my way out of a wet paper bag, I’m not a data scientist. I don’t know how many custom GPTs I’m really going to make. I still have to dig in with the tools I know about, much less trying to build cool new things. I do love throwing problems at various models though and seeing if they can do a better job than I could. Hey, when I figured out Copilot could read in an WordPress XML export file full of custom post types and fields and output something I could import like a plain old blog post, I was pretty amazed.

So it’s going to get bumpy. It’s going to get messy. There is going to be backlash when a rogue AI agent goes on an Amazon shopping spree with stolen credit cards. But it will pass. We’ll retire the stuff not ready for prime time, and then come out with even cooler stuff on the other side.