AI is everywhere now. It has become a genuine part of daily life, not a novelty, not an experiment, just infrastructure. And somewhere along the way something went wrong with how we talk about it.
Why is every AI company racing to announce how many thousands of jobs it is going to eliminate? When did we arrive at a place where the headline is not what we can do now, but who we can remove? Every company replacing humans with AI wholesale is doing it wrong, and most of them deserve what is coming.
A CEO with no engineering background can go into Cursor or Codex and prompt an app into existence. Fine. But he does not know when the model has hallucinated a dependency. He does not know which framework fits the problem, how to get it into production, how to set up Docker, who maintains it in six months when requirements change. The app gets built. Then it breaks. Then no one knows why.
AI does not have the creativity or the judgment of a human. What it has is speed and breadth. That is genuinely powerful, but only in the right hands.
This is how the new era should work. AI as an enhancer. A developer with AI can do the work of five or ten developers. That is extraordinary. The mistake is reading that as a reason to cut headcount. The smarter read is: now your team of ten can do what fifty could not. Most companies see the first option and take it. They reduce the team, keep the expectations, and wonder why everything starts shipping slower and breaking more often.
What actually happens is you hollow out the engineering org. You remove backend engineers, frontend engineers, people who hold context no documentation captures. The ones left inherit everything. Schedules get brutal. The bar keeps rising because AI is supposed to make everything faster. When AI first appeared in workflows it felt like breathing room. Time back. Time for the people and things outside the screen. Now, for a lot of people in engineering, it has made the job harder and the hours longer. Some of them genuinely wish it had never been introduced.
The layoffs will happen. That is the honest answer and dressing it up does not help anyone. But it should not stop you. It should probably do the opposite.
There are more problems visible right now than there are people solving them. An influx of new startups is trying to find them, and most will miss. But you might not. Everyone has access to the same models. Not everyone sees the same gaps. You might spot a sector no one has touched. You might build a UX so good the competition stops mattering.
The bottleneck is not the AI. It is your creativity. What can you do that no one else would think to do?
The Data
A large part of what AI actually is comes down to the data underneath it.
Here is how a transformer model works at a basic level. Pre-training is the first stage. You feed the model an enormous amount of text and it learns patterns across all of it, adjusting millions of internal parameters to understand how language fits together. It does not store the text itself. It distils it into weights. Think of it as absorbing the shape of language without keeping the words. At the end of pre-training the model can predict text but has no sense of how to be useful in a conversation.
Post-training is the second stage. This is where the model learns how to behave. Fine-tuning on specific tasks, RLHF where human raters score responses and the model adjusts toward what they prefer, constitutional and adversarial methods where other models are used to critique and correct outputs. This is where the personality, the safety behaviour, the usefulness comes from.
Pre-training is where the real controversy lives.
These models are trained on the internet. Your writing. Your artwork. Cookbooks, novels, source code, forum posts, academic papers. Copyrighted material feeding into a system you have no access to, no stake in, and no compensation from. That is a genuine problem and it is not resolved by the labs saying it is transformative use.
The bigger issue is what happens after. The model gets trained on the world's data, the result gets locked behind an API, and the justification is safety. Maybe some of it genuinely is. But every other major lab has released open weights. Meta with Llama. Google. Mistral built their entire identity on it. The holdouts train on data that was never fully theirs, keep the output to themselves, and call it responsibility. That is not a safety argument. That is a business model wearing one.
If you trained on the world's work, the world should have access to what came out of it.
AI and Devs
AI has changed the fundamentals of development. That comes with a cost too.
I am seeing it in my space. Layoffs. Developers who resent AI not because they do not understand it, but because understanding it has made their job harder and their position shakier. Covered the workload side earlier. But there is something more personal worth saying.
If you have spent five years grinding through code, the pain of not knowing what an array was, what a list was, when to use one over the other, building that instinct slowly and the hard way, and then AI comes along and a seventeen year old dropout builds PCIFIC, a full refurbished tech marketplace, in four months with it, that is going to feel like a waste. Like the ground shifted under five years of your life. That reaction makes complete sense.
But here is the honest view. Senior developers are the best people you can put in front of AI. It is not even close.
I know this because I am not one, and I feel that gap every day. I learned to code almost entirely with AI assistance. Instead of just taking the output I kept poking at it. What does this mean? Why does this work? When do I use this instead of that? AI is an extraordinary teacher when you actually interrogate it rather than just copy it. I picked up things fast.
But I also know what I am missing. If I pointed a senior developer at my codebase they would lock themselves in a bathroom and cry. They would come out red-eyed asking why is that there, why is it structured like this, what is this doing anywhere near production. And honestly I would not have great answers. I do not see those patterns yet. If it works it works, and that is a perfectly fine way to ship, but it is not the same as knowing why it works.
The new generation of agents is closing that gap fast. They are already writing better code than me, and better than a lot of professional output I have seen in the wild. We act like AI undermining professional codebases is some kind of scandal, but let us not pretend every senior developer was shipping pristine, perfectly considered code before any of this. They were not.
The senior developer who leans into AI becomes the person who knows when the agent is wrong. That instinct, built over years of pain, is exactly what the tool needs behind it. You are not redundant. You are the editor. The ones who refuse to engage at all are the ones who will genuinely struggle.