AI Needs Skeptics, Not Extremists
So what do we think of AI usage now in the middle of 2026?
This is an odd technology that has landed on our doorstep with a very loud thud, it has become a major facet of the post-lockdown world, right up there with inflation and hybrid work. It's useful for so many things! Researching ideas for papers, writing rote content, writing boilerplate code, identifying patterns and problems that more deterministic tools can't easily spot. I've found it particularly useful on technical documentation help. On the other hand it's also been useful for pooping out bland content and pretty questionable LinkedIn posts for engagement. But the quality is always good enough where its fairly engaging right up until a person realizes that they have been reading something machine-made. The code it produces is good right up until someone spots a major issue that calls the whole thing into question.
Things are moving so quickly that nuance is not taking hold very well. The extremists are coming out, grabbing their bullhorns and shouting into the masses to pay attention to their side. "We need to figure out how to use AI today before we get left behind!" versus "Any use of AI will be our downfall as a society!". I'm an opinionated devil's advocate, so both of these types of statements tend to rub me the wrong way.
I just worry that AI usage gets lumped into some mental zone where common sense statements go to live. And when they end up there we see lots of low mental energy aphorisms. "Don't use AI for anything important" getting lumped into the same category as "Don't throw out recyclables". It's a thought-terminating place to go. We are in too early of a stage of this thing and we need nay-sayers to help us cautiously wade into new workflows. We need people to keep experimenting with how to use it to really discover where the tools are helpful for whatever process we're applying it to. Blanket rules for AI usage - either "AI all the things" or "AI nothing" are simply not helpful at this place and time.
"AI will be used at every stage, I'm building a new production-ready AI bot that can make my job completely irrelevant." - Every other LinkedIn post in my news feed for some reason, I'm not sure what the goal is here other than trying to get a job at some cool-looking AI company.
Maybe the problem has been trying to apply AI too holistically. At first glance it looks like it can solve just about any problem. I mean, you can write your problem statements in English and the bot will poop out some pretty decent answers. Why not look at every part of your company and just say "What if we AI'd the shit out of all of this?" You can picture some Senior Executive getting teary-eyed over all of the value they're about to create for their shareholders. So by decree a company department sets along the task of putting an AI on their established processes and lo and behold, it doesn't really do anything of value beyond costing the company a lot of money in token usage.
Is this surprising? No. The executive didn't understand what AI could really do and neither did the implementers. It takes experimentation to find the value, and that takes time. It probably doesn't help that the assumptions one made about AI's capabilities months ago are no longer valid either. Just last year a chatbot could give you a very wrong but confident answer to a technical question. Today the chatbot will take its assumptions and put it through a few iterations of checking answers to make sure its right before presenting the final answer.
Most of the time in mid-2026 I find bots are right, like somewhere around 80%, but I notice now that they dull the answers to things they can't easily confirm. I was asking it to summarize some recent parts I had read of the Dungeon Crawler Carl series so I could refresh my memory of the events, and it gave a correct answer, but with zero details or examples. I realized it was probably grabbing some summaries from commenters or wiki pages and couldn't verify the source at all, so it just gave me some vague but true answers instead.
"Any AI usage is detrimental for both your brain and the environment." - A summary of many comments I've read over the past month.
Using the tool is now getting lumped into a similar space as smoking a cigarette, addictive and bad for you. I've seen a lot of talk about this in the last few months in particular: How AI usage is bad for our brains, the very use of AI is turning our minds into putty. Traditional writing is good for your brain, so is writing code, so is making mistakes and fixing them. These are engaging activities that take time and will have people stumble. The average person can't write 10 articles for an audience every day. The average coder can't write 1000 lines of code per day. But AI can do it, or at least it can do it at scale with mediocre quality. The high quality stuff takes time. But maybe AI can help out with some of those time-consuming parts - the editing, the inspecting, the thinking. And many would argue that offloading those tasks will dull your mind over time.
For this post I've been writing everything by hand, freewriting a lot and then rearranging and editing my thoughts. It's time consuming. I reached for Gemini to help me restructure the post into something more coherent and I found the output had that same veneer of what people identify as "AI slop". My content is probably going to be slop if I write it out by hand (I am not a good writer), but at least its my own slop that I can learn from. I wrote two posts in my blog early on with AI and they are honestly tough for me to read without getting extremely bored or annoyed with the style. It just doesn't sound like me or have much of my voice. So going forward I'm going to try engaging with the writing process by hand, but try to find where AI might keep my efforts in line or get me to the end product faster.
"Can we have the AI do that?" - A manager responding to some problem that was going to blow up the timeline for delivery.
The later half of 2026 will likely see the shift from "Put an AI on it" to something a little more nuanced. First we need a new frame of mind for applying AI. Almost all of us come from a scarcity mindset, so this technology seems very bizarre at first pass. Oh I can ask the AI about my particular situation involving the City of Seattle's building codes? I can even show it my letter from the city and it will know where to guide me on the website? Yes, but, it's also not 100% right about that answer, probably more like 80%, so you better pay attention. I think trying AI on various tasks is probably a good move, but just be willing to toss out its answer. I mentioned trying AI on writing this blog, and I found most of the result pretty useless. That's okay, the cost was negligible.
"You're not slopping enough." - A youtuber I watched recently who meant this will extreme sincerity.
That leads us to the latest development in the pro-AI movement: just do way more of it. The idea here is that AI is a post-scarcity technology that we can treat as such. If we can assume the output is good 80% of the time, and that 20% is the hard part, well let's just crank up the input and output to a point where that's not a problem anymore. So you have loop engineering and people saying that to be effective in this world, you should be maximizing your usage at every turn. Its brute-forcing AI usefulness. Its looking at the backlog and saying "okay most of this stuff AI can do for you if you give it a few passes."
There is some genuine value in this mental model, because at some point you need to start thinking outside the box to actually keep the AI slop machine moving in a positive direction. Instead of focusing on the primary task at hand, open up other parts of your workflow to it, maybe it can reduce that stuff to let you focus on the primary task. I'm thinking about how AI can build out many tools and tests for code. I wanted to record a video of a game I was making, why not have my bot create a tool that plays the game, records the video, and outputs all of the formats I need for sharing to social media? I honestly think that's a great development in our ability to explore workflows - just slop it out more and often.
On the other hand when trying to consult AI for writing, such as this very post, I found its output bland and uninspiring. It was only after asking it for detailed criticisms of this post, instead of re-writing my content, where I found the output really useful. It tells me areas to strengthen or improve or clarify, and I can figure out how to do that on my own. I think of that as having an old English teacher give me feedback - they typically would not rewrite my paper for me, they would point out the problems and make quippy suggestions. That improves my skills as well as the writing.
So brute-forcing output works for these internal tools where we can throw it out easily, but it falls apart when slop reaches the end audience. A consequence of having too much slop is that people find it too much to handle and they will ignore it. A good example would be the proliferation of Marvel shows and movies that followed Avengers: Endgame. Instead of giving the audience a chance to breathe after a momentous event and building up hype for a new cycle, Disney and Marvel pushed out many new characters and formats hoping to capitalize on their popularity. The quality declined, it became confusing to follow, and people voted to just sit it out and wait until something interesting came along again. So the same will go with too much slop for productive work - if we make a ton of tools around our workflows, it will get difficult to track them all and what purpose they hold. It will get difficult to track the value adds. And I can't even imagine a world of AI slop content to read or watch, it would devolve into bizarre and unwatchable so quickly.
I think the point of all this is that these tools are still in their infancy and judgement calls are still too early. The value is there, but where it has that sort of "word processors replace typists in the 90s" kind of value add is not well discovered yet. In tech we have found AI can spit out code at a ridiculous pace, and we are now inventing new ways to analyze the output or at least keep up with it. I'm still a bit dubious on it since discussions suggest there's more catch-up work to do to make sure the code actually works. If you've arrived at the conclusion that AI usage is a zero-sum game and using it at all will result in net negative effects, then I think you're missing out on the potential for actual improvement in your workspace. I won't claim that your competitors will overtake you or whatever, that's the same argument as FOMO, but I do believe that by terminating AI usage completely you're not really engaging in making incremental improvements or finding the specific areas that AI needs to be tamped down. We need objective nay-sayers, not religious ones.