So I made the mistake of opening LinkedIn, and reading into most posts there.
It was honestly depressing.
Not because of the market. Not because of the lack of good job opportunities. Not even because of the whole AI bubble – it was depressing because I realized that I could no longer tell if anything I was reading was real.
When the lies stop mattering
Some time ago, I watched a video (or maybe read a blog post, I honestly don’t remember) where a Russian citizen explained how misinformation affected people in his country; the things that stroke me as interesting is not that misinformation worked because people would believe the lies of the government, but what actually happened in the heads of the citizens. The video (or post) started by saying that, at first, the lies were almost believable.
Things like “We produced 10% more than last year.”; then it became 50%; then twice as much.
Eventually, the official economic indicators were pointing in one direction, while the government was saying exactly the opposite. The next step, obviously, was the government contradicting its own official information. Finally, people publishing the contradicting information were fired, because the regime always had to be right and everything always had to be getting better.
But regular people knew the truth. How can someone say “our goods are getting cheaper” when you pay more for everyday things?
Eventually, the lies became so obvious, so blatant, and so disconnected from reality that people became numb to them. It wasn’t that they stopped trusting the government and started believing the opposite…
They simply stopped caring.
If the government said it was winning a war, that information became meaningless. If it said ten people had died in some event, maybe ten people had died. Maybe a hundred. Maybe nobody. Maybe the event itself had never happened. There was literally no useful information anymore.
And when I opened LinkedIn, I felt something similar.
Is any of this real?
When I look at posts about AI, I cannot trust the positive posts.
I also cannot trust the negative ones.
I cannot see a trend on AI usage because I don’t know which information describes reality and which information exists only because someone wants engagement, investment, consulting clients, or a new job title containing the words “AI” and “strategist”.
For example, recently, I found a post claiming that a developer was producing 40,000 lines of code per week. Forty thousand.
This is completely insane.
The entire project I work on in my day-to-day job has approximately twice that number of lines. This person, this single person, is supposedly writing the equivalent of half of our project every week-or two complete copies of it every month. This simply cannot be real.
But then I saw a video from a known company, showing some internal project with millions of lines of generated code. So… is 40,000 lines per week real?
I honestly don’t know anymore.
Someone else said that they merged twenty, thirty, or forty pull requests per day. One person, merging more than what basically every team I worked with in my whole life merging that number. How?
Supposing you work eight hours a day (which you should, but for some reason our “hustle culture” feels it’s desirable to work way more than that) – 40 pull requests means five pull requests per hour – and we’re talking only about reviewing code, not writing it. Usually we need to understand what is being asked in a PR, then check lines, check styles, etc… if you take 5 minutes to review each PR…. we’re talking about half of your work being “review the code”, and the next half shared between understanding a feature, implementing said feature, opening up a PR, explaining on the PR what it does, and attending all meetings and other day-to-day work you need to do.
So, is “40 pull requests” a reality? Probably not.
But that raises another question: Why are people measuring this?
Meaningless metrics are still metrics
Every company measures productivity somehow. I am not saying this is good nor I am saying this is bad. I am just saying that, eventually, some kind of “performance” will need to be measured.
Now, suppose someone delivers four features in a week – this is actually a lot of features, but it is at least understandable. We can discuss what a “feature” means, how complicated each one was, whether they worked, and whether users needed them. Maybe “four features” mean “four small things that we need”, maybe a graph somewhere, maybe a code that invites new users, small things that can actually be implemented relatively quickly and without too much disruption.
Now suppose someone writes 500 lines of code in a week – what does that tell us?
Nothing.
Five hundred lines might be too many or too few – if I am bootstrapping a completely new system, for example 500 lines might be too little: the system needs to grow, reach production, and be validated as soon as possible.
If I am changing a legacy payment system that runs at a specific time every day, 500 new lines might be terrifying. I might be introducing a huge amount of complexity into something that should barely change. For context, I once wrote an invoice system – yes, it literally received data from clients, their line items (via a message broker), validated the “schema” of that message (with coercion, so an UUID string became a real UUID object in the language), then emitted the invoice in a specific format depending on the country, and uploaded that invoice to a specific system (again, depending on the country – SFTP for one, a RESTful API for other). The whole system had 3,200 lines of code, including tests.
When somebody says, “I wrote 5,000 lines of code today”, my first reaction is not admiration – my first reaction is asking: Why? Especially, why were you counting?
It is like someone working in construction saying: “Today, I lifted two tons of material for my company”. That sounds impressive until you think about it for more than five seconds: how many walls did you build? How many doors did you install? How many electrical systems did you finish? Did the building become any closer to being complete?
The total weight of the material you lifted says literally nothing about the work that was accomplished. Lines of code and pull requests numbers are the same.
Forty pull requests might be a disaster
Suppose you really merged forty pull requests in a day – what were they? Were they forty bug fixes? Because that might not be something to celebrate – did you really leave forty simple bugs unfixed so that you could fix all of them during the same week and publish an impressive number afterward?
How long had those bugs existed? Also, were customers affected? How many? How did these bugs appeared in the first places? Do they appeared because the previous forty pull requests made by LLMs introduced them? Also, did the pull requests contain meaningful changes, or did you split one change into forty tiny fragments because the metrics in your company rewards quantity instead of quality?
A number without context can describe success or complete failure – and yet, people boast about these statistics as if producing more code were obviously good.
For literal decades, we have discussed how less code is usually easier to understand, test, change, and maintain. Reusing an existing abstraction is generally better than duplicating it. Deleting unnecessary code can be more valuable than adding a new module. Then AI appeared, and suddenly producing an absurd amount of code became a superpower.
The free guide to an impossible advantage
There is something even stranger about these posts – people don’t simply say:
I can merge forty pull requests per day.
They say:
I can merge forty pull requests per day, and you can too. Here is my guide…
Usually, the guide is free too, a post, or a very cheap “crash course on AI”. Let’s suppose this is real – suppose you discovered a technique that allows you to produce forty useful, correct, reviewed pull requests every day (I’m not even mentioning “hour” because that’s too far from reality in my opinion). Your company considers this valuable. You consider it valuable. It supposedly makes you dramatically more productive than every other developer.
Then why are you publishing it on LinkedIn? Sure, maybe it is something so simple and obvious that everybody can do it. Maybe it genuinely gives you an enormous advantage. In that case, why would you immediately give that advantage away?
None of this makes sense.
I remember hearing a similar argument about companies selling AI tokens – the argument came from someone I very much do not admire, but unfortunately, a bad person can still make a good point occasionally. He mentioned something like:
If these models can generate such an absurd amount of wealth, why are the companies selling access to them? Why are these companies not using the “tokens” to generate the wealth from themselves?
It’s like you have a completely legal machine that generates money, without inflation, legal problems, or any other consequence. Literally makes money appear from thin air, and you keep the money, can use it at will, and it is not illegal – you literally invented a money-making machine that can make you insanely rich. Why are you selling tickets that allow other people to use the machine? If that was true, you would not do that:
You would use the machine. You would generate the money. That’s it.
And yet, companies supposedly possess a tool capable of replacing entire departments, creating whole products in minutes, and producing unlimited value-and their business model is to sell the access of that tool to other companies.
Sure, maybe the tool is useful – I believe it is useful for some things. But that is very different from believing it is a magical wealth-producing machine.
Better than Kryptonite
People are terrified of admitting this – AI can be useful without being a superpower.
It can help with some tasks and fail horribly at others. It can generate a good first draft and a completely broken second one. It can save twenty minutes today and waste three hours tomorrow. Equal prompts can generate very different results, one being insanely good, and the other being insanely bad.
But that is not how people describe it on LinkedIn – On LinkedIn, ChatGPT is apparently more powerful than Kryptonite. If Superman existed, someone would publish a post explaining how they defeated him with a simple prompt-without even using Kryptonite-and then sell a course explaining the exact prompt.
And the comments would contain hundreds of people saying:
Game changer.
This is the future.
Developers who don’t adapt will be left behind.
Meanwhile, nobody even ask themselves if Superman actually exists, if the fight happened, or if the prompt did anything.
Swimming in the sea of misinformation
When I see people making claims about AI, but I don’t know whether those claims are real. I don’t know whether the people are real. Sometimes I don’t even know whether the companies they supposedly work for are real – and after enough of this, I start doubting whether my own area exists.
The software industry described on LinkedIn is completely disconnected from the software industry I experience.
In one, people spend days debugging strange production problems, discussing trade-offs, trying to understand old code, reviewing changes, talking to users, and discovering that a supposedly simple feature depends on seventeen things nobody remembered.
In the other, someone types a sentence into a textbox and produces an entire company before lunch.
Sometimes it feels like watching a group of people planning a castle that needs to withstand a cannonball – they are debating whether the castle should be made from glass bricks or plastic bricks. Both options are obviously too fragile, but that is not even the biggest problem:
The problem is, the glass bricks do not exist. Nor the plastic ones. What people are discussing is which imaginary, fragile material they prefer – then they will commission a new company to invent and manufacture the bricks. Then they will build the cannonball-resistant castle.
But the deadline is in two weeks, and everyone in the meeting considers this a reasonable
Sometimes I feel that I am dreaming; or maybe this a nightmare.
I can’t believe that this is simply how business works now – that we publish increasingly absurd claims until nobody can distinguish a real result from a completely fictional one, repeat those claims until investors provide money, and when the money is gone and the claims can no longer be sustained, we disappear and leave everyone else with the debt, the unfinished systems, and millions of lines of generated code that nobody understands (or even exist, but the CTO swears they do, they are just deployed somewhere but the last deployer was Marcus that left the company last year, although the employees’ records don’t show any Marcus working in the company).
And that is exactly the problem.
The information has become so exaggerated, contradictory, and detached from reality that evaluating whether a specific claim is true no longer feels useful. Sure, maybe someone really did generate 40,000 lines of code this week; maybe they did merge forty pull requests today; maybe an AI built an entire application while the developer was drinking coffee.
Or maybe none of it happened.
And what scares me is that all of those possibilities feel equally plausible.