Every single day that passes inside the current AI bubble somehow manages to do two things:

  1. Make me feel like I must be insane because I don’t get it.
  2. Remind me that what I see is so stupid that I’m definitely not insane.

It’s hard to reconcile these two feelings constantly.

I hope we’ve basically reached peak AI madness because I have no idea where else the stupidity can go. Let me give you a few professional and personal examples:

MOAR!

I am regularly subjected to documents written heavily by AI. This should come as no surprise that these often come from Product departments.

Aside: It Has to be Said

You know what? Let me hold off on AI mockery for a moment to unleash some amount of fury on arguably my most despised department in modern corporations: Product.

While I have worked with some excellent individuals in my career who work in Product, they have always been excellent in spite of their useless and expensive department. In fact, these individuals have excelled typically because they were willing to immerse themselves in some other aspect of the business (sales, R&D, operations, etc.) or act as a shield between executives and other suits who believe pulling a timeline out of thin air qualifies as reasonable.

Aside from these very few individual exceptions, Product is a sort of platonic ideal of the corporate parasite. Sales and finance are full of middlemen. Product is what happens when someone decides that creatives and makers also need middlemen to suck up time, energy, and salary budget.

Let’s take stock of who typically fills these roles:

  • MBAs (ouch)
  • inept designers
  • failed engineers
  • a younger generation of assholes who have actually decided to just be “product people” (there’s something to pass on to your grandchildren)

Now, let’s have a look at what they produce:

  • Powerpoint presentations.
  • Meetings… oh so many meetings.
  • Some new fucking planning paradigm that’s as dysfunctional as the last because these people aren’t typically involved in any actual sausage making.
  • Incomplete and under-researched product documentation because it’s hard to plan the details of sausage making when you either have no idea how or sucked at it when you did. These documents are now tarted up with AI—and by “tarted up” I mean mostly shit out by AI.

Back to AI Sucking

Anyway… these goddamn AI documents. I hate them. Whenever some fucking asshole starts a meeting by sheepishly saying something like, “Sorry, I had to put this together in a hurry with the help of ChatGPT,” I want to throw a chair. Let’s translate that shit for a moment:

Hey, I’m much too busy with stuff that’s more important than everyone else in this meeting to even do the bare minimum of my job. I have summoned an AI chatbot to create a document that, while almost entirely useless in practical terms, looks mostly complete and can go on a huge stack of papers that executives—because they are such simple-minded creatures—will mistake for real work. With that out of the way, I hope everyone else here can fill in the gaps—which means producing what should have been in this document before this meeting started.

Frankly, it’s disrespectful. It’s also standard operating procedure where I work these days. The most irritating thing is I don’t get the impression the people producing these documents actually believe they’re particularly useful. If they did, there would be no need to apologize. When they don’t apologize at the outset, what typically happens is I’ll notice glaring factual errors when reading the document—for instance, references to vendor APIs that do not even exist—and ask if I’m insane or blind and get answers like, “Oh sorry. ChatGPT is hallucinating again, lol.”

Really? Well that’s fucking great. Now I have a document that was supposed to, in theory, have been compiled by the department that was supposed to do the initial research and coordination for me but now I have to review every single fucking word with utter distrust. So while ChatGPT may have allowed you to get things done faster, it’s going to cost me considerable time to get someone else’s work in the state it should have been before I was even taken away from actually building shit to review.

What did AI do for anyone here except create a document that looks very complete provided it is subjected to precisely zero scrutiny? The only upside, if it can be called that, is that it has laid bare how much of this “process” and the people behind said process could just evaporate one day and things would still function—and probably function more efficiently.

The kicker, and really the reason for even writing this, is when these AI generated maelstroms are unleashed on makers who are too busy to keep up with the insane volume being thrown at them because Product can now generate no value at 10x the previous rate, some engineers resort to… asking AI to generate a summary of the bloat already generated by AI.

My guys. My guys. What the actual fuck is even going on? Work and production are not measured by volume. In fact, the best production is where the volume to cash ratio heavily favors cash. The reason software became so huge over the last 30-40 years is because it can be replicated so easily. Who decided it was a good idea to suddenly produce more useless paperwork than contract attorneys?

Again, what the actual fuck is even going on?

Also, my dear corporate overlords and C-suite elite, aren’t you supposed to be a bunch of profit-driven amoral capitalist ghouls? Don’t you realize none of this is making you wealthier? Can you please use your vampiric greed for my benefit just this once and let me build the shit that improves your stock position? Good lord. Instead, you slapdicks are giving each other high fives while glancing at all the paperwork as if it is the sum of production.

Spoiler alert: it’s not.

TL;DR: People are generating documents using AI by providing prompts that are little more than summaries and then other people are using AI to summarize the expansions of those summaries.

The question must be asked—in case I haven’t yet—what the actual fuck is even going on?

AI writes. AI reviews. What the fuck do you do?

I hate AI (more specifically I hate LLMs). I realize this is obvious, but I have become a full scale hater at every level. My eyes dart right past the AI summaries produced by search engines because fuck you, that’s why. While I can’t entirely avoid it as it metastasizes to like… everything, I can do the one thing any individual can do: just say no.

So here it is, no. No, I’m not going to try it anymore. No, I’m not going turn it on in my text editor. No, I’m not going to ask it a single fucking question. No, I’m not going to entertain it at all. No, it’s not a tool in my tool belt because I have too much respect for what I do and the sustainability of what I leave behind.

I’ve yet to see a single thing AI produces that is faster than what I do on my own, especially when measured over any useful period of time—and this distinction is key. I can’t make a chat app in a couple minutes, but who cares? If I bothered to write a chat app, it would need to designed for my business and specific needs and that requires all that thought and planning stuff. Otherwise, you know what is just as fast as shitting out another chat app with AI? Integrating some existing chat app or library. Do you have AI write a JSON parser for you? Only if you’re completely insane.

And let me be clear, I am not some kind of code god, okay? AI code generation simply cannot do the hard stuff. It can’t design. It can’t plan. (No, it can’t. Shut up. That’s not planning.) It can’t understand a system’s present or future needs. It can’t calculate necessary compromises. It can’t write documentation that’s useful.

Despite all this, it has not stopped the least skilled programmers from spreading this cancer throughout the land. Mountains of code is written by AI. The tests (if there are any) are written by AI. Junk documentation is generated from the generated junk code. Junk commit messages are summaries or junk documentation and junk code. People who are lost in all of this ask for code reviews from experienced members of the team who pad their bullshit metrics by asking @claude to handle it and then call it a day for their in depth “review.”

Hey, you. Yeah, you because if you’re reading this you know who you are. Fuck you sideways. You do your profession, your coworkers, yourself, and your species a huge disservice. Please, I implore you, stop being both a lazy slop architect and an asshole.

Who is the person who really thinks having a probability engine incapable of any actual thought “design” and sustain systems is a good idea? Oh yeah, probably the most senior dipshit in your department because this person has long since detached themselves from any semblance of being an actual maker. This person mostly just looks at reports and spreadsheets now and what does this person see? By god, this person sees volume. This person knows this is the year AI is transformative. (Just like 2023, 2024, and 2025…)

By the way, this person is right. If I’ve made no other point beyond the fact that Product departments suck, it’s that we’ve reached peak AI. We have AI summarizing and reviewing itself. While this is quite transformative, it’s fucking horrible. It has absolutely transformed our ability to generate more and more volume. Just… more. More and more and more and more.

In case you didn’t know…

If you’re not a programmer or you’re one of these so-called prompt “engineers” you might not be aware of something: good programming practices tend to revolve around reducing volume. Things like DRY (Don’t Repeat Yourself) and SRP (Single Responsibility Principle) are both bedrock concepts for good design and also inherently at odds with the slop AI produces.

AI improving your workflow is a sign that volume is already a problem. If I hear one more goddamn programmer say, “Well, it handles the boilerplate,” I might just end it all for at least one of us. You know what else handles boilerplate? Tools we have had at our disposal for decades such as macros, meta-programming, purpose-written code generators, and all sorts of other goodies that you might be aware of if you didn’t treat your craft like amateur hour!

How long does “boilerplate” take to handle the old-fashioned way anyway? Seriously. If this is just such a huge part of your workflow, what is going on over there? What? I legitimately struggle to think of a single time in the last decade of my work that I’ve found myself thinking, “God, all this boilerplate eats up so much of my day. I need to tackle this head on and get it under control!” I spend an order of magnitude more time agonizing over naming modules and functions than I do with anything even resembling “boilerplate.”

This kind of repetition always speaks to much, much bigger problems either in your own abilities or in the codebase. Guess what? Copy pasta bullshit isn’t going to make this problem go away. I’ve already written about this anyway.

Years ago, I came across a blog entry titled Kill your Dependencies by Mike Perham of Sidekiq fame. While I recommend you just go read the whole thing, I want to quote his “Rules to Remember” directly because in the almost decade since he wrote them, they have become an integral part of how I review my own work:

  • No code runs faster than no code.
  • No code has fewer bugs than no code.
  • No code uses less memory than no code.
  • No code is easier to understand than no code.

This opened my eyes to a terrible truth: every line of code is a liability. (Out of the Software Crisis later hammered it into my head even harder.) It may be a necessary liability, but it’s a liability nonetheless. Again, the way AI generates code is inherently at odds with how code should be generated.

You could argue—as every single AI touting buffoon out there does—that code generation is meant to have human oversight. And you’re right! If you’ve ever worked in a reality where people actually believe this stuff is turning them into 10x developers, you’d know this is exactly not what happens. Oh sure, humans do have to step in, but the goal is for them to sully their hands as little as possible.

I have not come across a single programmer who makes heavy use of AI and treats it like an actual junior developer, meaning there is a heavy review process and then major design optimizations. This would result in volume being reduced through basic good practice. The reason this is rare, if it even exists at all, is that anyone who knows how to use a stopwatch is going to very quickly realize that not only was no time actually saved, but arguing with Claude and then doing proper cleanup is probably less useful in keeping their own skills sharp.

If you think your mind and your ability to create are static and don’t require constantly maintenance and exercise, you’re in for a nasty surprise when this bubble bursts. AI usage encourages programmers to put forth absolute minimum effort under the auspice of “only doing the fun stuff” or “focusing on what’s important.”

I could write an entire essay on how fucking wrong this mindset is in the first place but I’m not sure there is a reason because both statements are pure fiction anyway!

Oh, and when did reviewing someone else’s code become the fun part? What is that? I thought the sales pitch here was that I got to focus on the fun stuff. Good reviews take time. People put them off all the time because almost no one wants to do them. When I review design work by juniors, the review can take longer than if I had just done the work myself whenever complex design or domain modeling is at stake. The review isn’t for my sake though, it’s for the sake of developing juniors into something greater and cultivating their skills and knowledge.

When you’re on mop up duty for Claude or Codex or whatever LLM you’re jerking off, are you really reviewing everything? (We are men of action. Lies do not become us.1) Is that process really faster? (Not if you were reviewing everything.) What’s more, if you give me the line that, “it’s like a junior developer” are you growing as an engineer just reviewing this crap?

I think that’s what disgusts me so much about AI in code and its veneration. It can only magnify and spread existing problems. If your team or organization has embraced this methodology not only is it producing more and more slop2 but it’s stunting the development of the most valuable part of the system: the actual fucking programmers.

Also, if this shit is so good, where’s the shovelware?

It’s great for prototypes though!

No, it’s not.

If I have learned nothing else in my career it’s that the prototype becomes the product. On the list of lies we tell ourselves or are told by our bosses, treating anything we build as “throwaway” is among the worst and most frequent.

In actual engineering, throwing away the prototype has to happen because of the nature of mass production. The construction of prototypes is very different than rolling out the assembly lines for the final product. In software, there are no such realities and so “we’ll just launch with this and fix it later” or “after we prove that this works” and other variants take quick and dirty hacks and make them the core of some system.3

Oh, what’s that you say? You don’t have a boss? This is for a side hustle? Okay, so… how is this actually any different? Should this prototype prove to be successful and suddenly your one man shop has to support it, expand it, manage the business, and do all the other things that even modest success forces, you’re totally gonna have the time to revisit that… uhhh… prototype that totally isn’t your business now, right?

You gotta pay the piper. There’s just no way around it. No Silver Bullet is as meaningful today as it was when it was written. If whatever you’re working on isn’t meaningful, then go ahead, use AI. That begs the question though, what are you doing with your life?

But, but, but… most of programming isn’t even code!

I have heard this one a lot, somehow, as a defense of using AI. It’s all under the idea that AI somehow frees you up to do the important things. While I agree that code itself doesn’t consume the majority of my professional time because thinking, planning, documenting, and communicating will consume more in total, I don’t see how this is a feather in AI’s cap at all.

If coding isn’t a majority of your time in the first place, how on earth is optimizing that part of the process going to make you 10x anything? If code is less than a majority of your time in the absolute best case scenario an LLM can’t even make you a 2x developer.

Do you really want to argue that a chatbot can synthesize everything in your head better than you can, all while translating human language to abstract concepts and then into code? That says way more about you than it does about AI.

Summaries of Introductions

Okay, this isn’t code related at all. It’s not even in my professional sphere. It is, however, about more needless volume.

SEO has already done its very best to make finding content on the World Wide Web that’s not Wikipedia pretty awful. Even worse is how it’s transformed many, many blogs. AI is apparently the next step in the descent toward rendering most the information available on the information superhighway frustrating at best and misinformation at worst.

Ever come across a blog article that “summarizes” an introduction paragraph with bullet points immediately following the paragraph? And somehow the bullet list is capable of being longer in some instances. It’s sort of amazing in the worst way one could be amazed.

I don’t think I even need to comment on all the AI generated slop everywhere else. YouTube? All social media? Just everywhere.

Why does anyone believe software generated this way is somehow not a victim of all the exact some problems?

On the Topic of Summaries

Hell, even the summaries that I used to think were useful are still so fraught with mistakes that they can’t be trusted. Reading Gemini notes of meetings has really put into focus just how much of this is pure probability guesswork. It’s just more data of questionable accuracy that no one will ever use ever again.

I’m sure building LLM-powered summary bots to comb through all these poorly summarized notes is going to be extremely helpful. It’s so weird, but my notes sometimes written in pencil on actual paper during meetings are more useful to me because—and I know this is crazy—I know what I need to prioritize!

Why Analog Is Better

I didn’t have a plan for a conclusion. The summaries or AI generated documents plus my general work frustration of dealing with a bunch of Claude monkeys has been weighing on me a lot lately.

I’ve spent more and more time away from code beyond my day job and find it harder to take anything seriously.

I’ve recently gotten into the basics of audio engineering because I’m the bass player in a band and audio is… you know, an essential component in music. While looking for an explanation on the difference between gain and volume, I came across The Real Reason Why Analog Recording Is Better. Even if you’re not into audio at all, I think it’s a really good video sort of generally related to all of this.

Just watch it—and watch it at 1x please. Stop and watch and think and feel without life-hacking for 12 minutes and 18 seconds.

Process matters. Process matters a lot. It’s hard to quantify when looking at components in isolation but so obvious when looking at large projects and complex systems. That’s why it’s so elusive to the intellectually myopic.

  1. If you don’t know the reference, educate thine self! ↩

  2. Hey! In the event you didn’t know, slop is typically associated with the garbage you’d feed to pigs on a farm. Really think about that. ↩

  3. This is so pervasive you almost can’t not believe it. I have worked under multiple CEOs who think there is a simpler slider between delivery time and quality. The must crushing part is years in this same fucking guy is asking, “Why does everything take so long?” Crippling interest payments my guy… crippling interest payments. The full circle is apparently impossible to comprehend once you reach the inner circle of the C-suite. ↩