Wow, what a ride the last few years have been. 😬

The Amish aside (straw hat tip), it feels like almost nobody on the planet can escape hearing about AI multiple times a day at this point, for better or worse. Mostly worse.

As someone that’s gone in hard with this stuff since I first tried out DALL·E 2 in 2022 (and has since explored countless tools and agents), I wanted to finally dust off this long-in-my-drafts-folder piece and share my personal hot takes on the good, the bad and the ugly surrounding LLMs.

Caveat: I’m an American, so this’ll be centered primarily around the United States, but I’m confident that the wild-ass vibes right now are fairly similar no matter where you live.

I’ll start by saying that I’m explicitly not going to dive into the environmental, intellectual property infringement and “uh, it might kill us all” aspects of this technology. All are obviously problematic/terrifying, but have you…met our country?

We’ve seen unapologetic land grabs before and we’ll see them again, and while there’s never been anything even close to this magnitude, I’m fairly confident that though many of these volatile AI companies may not be long for this world, the technology itself (warts and all) is not going back in the bottle.

So with all of that hand waving out of the way, let’s talk about what this stuff is and isn’t good for right now, along with what it’s already irrevocably changed.

I’ve broken things down into a few use cases, some of which may not be as familiar to those outside of the technology industry. Let’s jump in…

AI Chatbots

This is the primary way most regular (non-tech industry) people experience LLMs, and it’s been the source of some of the most problematic experiences, which can range from flat-out misinformation (“hallucinations”) all the way to encouragement of suicidal behavior.

Not great, though things are steadily improving on this front, especially with more and more bots that cite their sources to assist with verification, and ongoing safety improvements as new models emerge. The larger issue though, especially here in America, is that many people love nothing more than having someone (or in this case something) very confidently just tell them what to do. Full stop. When we reduce decision making and advice seeking down to a simple and confident “here’s your answer!” framework from on high, things never end well.

Finding Information

We all mourn the loss of the non-AI-ified Google, but let’s be honest, for a huge majority of searches, AI search results (when accurate) can be pretty convenient and time saving, and this is clearly where things are heading. I’ll also say that while it’s slightly terrifying to see everything that these companies have on me after years (or decades) of usage, the ability to search across the vast corpus of your own information (via the new Siri for Apple users or Gemini for Google users) is flat-out gamechanging.

What I very much don’t like is how this dynamic shifts eyeballs away from those publishing on the open web to those that control these giant portals.

Creating or Viewing AI-generated Content (Text, Images, Videos, Audio)

All of this stuff is mostly still terrible, and obviously there are existential questions about how these tools will affect professionals in these fields, but here’s generally where I come down on their use:

• When desktop publishing first emerged in the ’80s (or HTML editors like Microsoft FrontPage in the ’90s), there was a wild initial burst of interest/usage as people felt empowered to create something for the first time, despite not having formal training (more on this below re: software development). Most of the results were of course absolute garbage, and we soon returned back to hiring professionals when it mattered, but I’m never one to shit on people trying to be creative. Let millions of flowers bloom! Even the terrible ones. Especially the terrible ones!
  
  
• Don’t sleep on the comedy value of slop creations. We live in dark times, and sometimes an absolutely insane photo/video/song is just what’s needed to give you a little boost. This is pretty much exclusively what I use AI images for in my Junk Drawer and Vitamin (C)haos newsletters. Never unironically.
  
  
• All that said, we absolutely need to keep pushing to support the emerging “created with AI” label standards, since there’s clearly a very dark side here re: misinformation and abuse. In other words, AI-created content should always be presented (and mocked) as such.

Weaving AI tools and Workflows Into Your Work Life

This is where things get fraught as hell. There are tons of no-brainer daily uses for AI tools, including recording/transcript tools and cross-service agentic integrations (via the MCP protocol), but things immediately get dicier as soon as someone starts to lean on AI to do their thinking and/or document creation.

The term “slop merchants” (those that generate something with AI and send it to another human without much review) has been flying around a lot recently, and I find it to be quite apt.

Why? Well, what I’ve seen professionally and personally is that AI tools can be a massive productivity boost for those that are already well-versed in the topic at hand, but a massive liability for those that only know enough to be dangerous.

When you create a document filled with garbage and obvious hallucinations, you feel great about checking that task off your to-do list, but you a) look like a fool to your recipients and b) have just created more work for the poor bastard tasked with cleaning up your little travesty.

You need foundational knowledge first, and only then does layering AI on top help you and others move faster and be more effective.

Which leads me to…

AI in Education

I’m married to a high school English teacher, so as you might imagine, she has thoughts on this topic.

Mine strongly mirror what I’ve witnessed (per above) in the professional world. You need to have mastered the basics before attempting to take things up a notch. Outsourcing tasks like writing and art to the machines a) kills your personal voice before you’ve had time to cultivate it and b) makes you forever dependent on these tools.

AI can be great as an additive, but tragedy strikes when it’s a replacement.

Software Development

This area has been the wildest to watch. I’ve been working professionally in this world for 26 years now, and I’ve never seen a tsunami of change happen as fast as the events of November/December 2025 (when Claude Code really came into its own) through to today.

Not only do the vast majority of software engineers barely write code themselves anymore (primarily now acting as editors vs. writers), but the “vibe coding” movement has blown the doors wide open in terms of who can now create functional software in literally minutes.

Is any of it perfect? Of course not. Should you build something wildly ambitious without consulting a real engineer or security expert? See above re: learning the foundations first. Should you jump in and build something small and reasonable as a first-timer to learn the ropes? ABSOLUTELY.

As someone that built websites back in prehistoric times, but then fell off the wagon as bloated frameworks and overly-complicated nonsense infiltrated everything like an invasive species, being able to just build and bring your idea to fruition is incredibly liberating, and you can totally feel the “early web” energy in the air.

Less doomscrolling and soulless product rollouts from giant companies, more learning, tinkering and launching your own stuff!

Parting Thoughts

While we often hear from AI-pilled maniacs on one side (LinkedIn and Silicon Valley generally), and “I’ll never touch anything related to AI, and should you feel personally bad for using any of it” crowd on the other (Mastodon, Bluesky, etc.), I’ve found that while there’s a great deal of stinky bathwater surrounding this technology, there is indeed a real baby in there too.

We humans contain multitudes, and making reasonable use of/staying curious about a technology that is truly gamechanging/liberating, while also loudly advocating for reasonable reforms and safeguards feels to me like the most viable path forward.

If I’m wrong, we can always fall back to Plan B: send someone back in time to kill Sarah Connor.