AI Coding Is Making Devs Miserable (2026 Predictions)

Our first prediction is that AI won't take your developer job. It will just make it shittier. On SVE bench verified, it seems like Cloud Sun can do up to 80% of the task that a human developer is able to do. But private benchmarks reflect a way worse performance. This is SVE Bench Pro, Cloud Opus performance drops to 45%. And if we expose it to private data sets, we see that performance goes down to 23%.

The LLM it's only able to solve one in four problems. There's an obvious reason for this. LLMs don't think, LLMs don't code. They just predict the next token. What LLMs are at their core? It's different layers of neural networks and you train this neural network on huge amounts of text. The two data sets that you see here is the fine web, which is basically all the data the internet has to offer. And then we have the big code, the stack V2.

This is programming data, so you feed all that into your neural network. Then this neural network will learn the statistical patterns, the relationships between your data. And what you have at this point, it's a very powerful auto complete. When you ask it a question, what the model will try to do is to output what's the most probable token that will actually fit the sequence. This probabilistic nature of LLMs has a very serious limitation, and that is hallucinations.

So when you ask the model a question, it doesn't know. What the model will do, it will not say, I don't know, but what it does, it it will look into this mathematical space of tokens. And it will say, oh, you know what, um this is the next the most probable token, even if this is not the right token, the right answer for the question. It will give you the most probable one. So for example, you ask the model, hey, why did John Cena became ambassador of to Spain, and the model will come up with with an answer.

That seems seems fine, but it's total BS. The first fix was actually giving the model access to tools so it can check its work. model access to the internet, access to your code base, you can also instruct your model better, what's called context engineering. For example, specify the style of your code, the files to look at, etc. Problem is, and this is kind of counterintuitive, that sometimes the more context you use, the more information you give to the model, the more the model can get confused and make mistakes.

And LLM companies are well aware of context limitations. Uh this is from Cloud Code Docs. Uh and it says, hey, this matters since LLM performance degrades as context fills, the more context we get, the less the LLM will will perform. When the context window it's getting full, Cloud may start forgetting earlier instructions or making more mistakes. If you want me to make an in-depth video on how LLMs actually work for software engineers, then comment LLM below.

The point I'm trying to make, it's that you can over engineer LLMs as much as you want. But at its core, what you get, it's a token predictor that makes constant mistakes with no understanding of what's actually predicting. As long as the underlying technology, it's an LLM, hallucinations won't go away and these models will keep on making mistakes. Even Andrea Carpathy himself said, hey, don't treat these models as infallible models.

Check their work. Use them as tools. Use them for inspiration, but work with them as tools and be ultimately responsible for the product of your work, which is exactly why AI won't take your developer job. I like to call it the verification gap, because just like firefighters, tourists and notories and power plant operators, developers will get paid just to be there in case shit happens. Even if AI is rating 95, 98, 99% of the code.

Meanwhile, AI coding over us is turning the software industry into a massive dumpster. What's happening is that software engineers are being forced to become professional slop reviewers. For example, game engine God, it's drawing in AI slop contributions. understand the code they are submitting. Developers have to deal with co-workers creating massive, useless PRs. But one of the bigger problems is that people shipping AI code are putting pressure on other people to also ship AI code.

Like this developer who said he's gotten to a point where I cannot not do it, not use AI, because then I look slow uh compared to to everyone else. I feel like every day at the office, it's a waste to see who can generate more PRs with AI and blitz stories. Whenever I think about the modern software engineering life, this image of Wally came to me, where we went from the people building things and thinking about things to the ones being forced to make sense and recycle whatever AI slap uh the LLMs are coming up with.

But there is some light at the end of the tunnel. We see this in open source projects, this is uh the ghost project, which has some very strict rules for AI usage. For example, the human in the loop must fully understand all code and people who produce bad contributions that are clearly AI slap will be added to our public denouncement list. Honestly, folks, I believe this is the future for both open source and non open source software.

My tip, it's, hey, until management wakes up, you will have to put up with AI slap PRs. Get good at code reviewing, code by hand, without AI as much as you can. Prediction number two. Thanks to AI, this year will go down in history as the year of technical depth. Because, hey, now we do vibe coding, where twin engineers can now create the tech depth of at least 50 engineers. Git clears 2025 report shows an eightfold increase in duplicated code blocks.

That means eight times more freaking dead code. And my question for you folks is, what happens when every website looks the same? Because every product I'm getting my hands on this console on the Atari 2600, everybody could write games for it, and people started shipping tons of games, uh buggy games, games that didn't really work. So the video game market actually crashed in 1983, which led companies like Nintendo to create a seal of quality, which would actually guarantee that the game has been tested and that the game meets certain level of quality.

As consumers get sick of shitty software flooding the market, they will demand quality. And of course, this will push in the the industry and companies back to a quality before quantity rule. The developer community will follow. Which brings us to prediction number three. The dev community will get back to the fundamentals and Vibe coders will be replaced by purists. Because when the focus gets back to performance, gets back to quality, gets back to reliability, the people that are just out there vibe coding some sloppy piece of SAS and trying to get some money or some attention uh from the user, they will get pushed out of the market.

And honestly, I cannot wait for the good old times to be back. Because before this AI paranoia began, we used to talk about things like the event loop. We used to think about how to make software better, we came up with things like React, things like functional programming in JavaScript. And every since this paranoia kind of took over the software industry, we're not really talking about anything else.

Prediction number four, the AI bubble is going to pop, thank God. Now, AI CEOs will keep on lying so they can raise money like Dario Amodei going on every podcast and every event possible and saying, hey, you know, six months from now, there there won't be any developers. 90% of the code will be written by AI, blah, blah, blah, blah. The same old story. Or Nvidia CEO saying, hey, AI boom is fueling.

The largest infrastructure built out in history. This is BS folks. GPUs have a very short lifespan. These are not internet cables that are going to be there for a few decades. Meanwhile, AI slop is eating the freaking word. in your average Google search looks like this, where almost half of the stuff that pops up in the search, it's actually fake. But soon the money will run out and this mass delusion will be confronted with reality. The emperor has no clothes, folks.

We've already seen that a few days ago with Oracle layoffs, tech giant slashes 30,000 jobs as banks pull out from financing AI data center. Now, open AI it's probably going to burn about 14 billion in 2026, hitting bankruptcy by next year. And because of the nature of this circular financing, which is like all these companies are giving money to each other, when one of them fails, there's going to be a chain effect, right? There's going to be a chain reaction and the whole ecosystem will implode.

My advice is find a job in an industry that is not directly dependent on AI money to survive and stick to it until the whole thing pops. Prediction number five, companies realize AI it's an empty promise, they will start hiring again. They have job market recovers. If you look at the data from last year, you will see a slow but positive trend. There are more and more job postings. Right, we've seen this happening in Canada, we've seen this happening in the US.

All in all, the job market it's slowly getting better. And it will keep getting better as tech companies are already regretting having laid off so many people. We've already seen this tendency in our mentorship program where in the last few months there's been a huge increase in uh people getting offers. By the way, folks, if you're a JavaScript engineer and you want to get to senior, Bogdan and I built the software mastery mentorship exactly to help you achieve that. You can see more about how we can help you following the link in comments.

Prediction number six, junior dev hiring will start picking up again. We know that because junior level jobs were not killed by AI, they were killed by high interest rates and by the economy slowing down, but it seems like the tide is changing and IBM, for example, it's tripling the number of Gen Z entry level jobs after finding the limits of AI adoption. All the AWS CEO is explaining the three reasons why AI can't replace a junior devs.

junior dev, believe in yourself. Don't listen to the noise. Don't listen to the grifters. Don't listen to the people telling you that your skills are useless, that you spent four years learning for nothing. that your GPA is useless. Nothing can be fighter from the truth. In this age of AS slap everywhere, companies will be desperate for people who can actually use their brain. So use your brain, stop overusing AI and do the work.

Code by hand. Prediction number seven, developers will split into classes in the long run, only one will thrive. We've seen some people or specifically junior developers who can't actually code because they depend on AI. This doesn't only happen at junior level. Many senior engineers are doing the same. My advice to you is to become the developer who can actually deliver value even when your company it's out of tokens.

Even when cloud code it's offline. Prediction number eight, live coding becomes the main filter in technical interviews. And this is from a survey to thousands of HR firms. It turns out live coding interviews were the third most commonly used method to assess skills and within those live coding algorithmic questions were still number one. At the time where most developers are forgetting how to code.

And here is my tip for you. Live code as much as you can, first in TypeScript or JavaScript, then move towards lead code. Prediction number nine, job hoping to increase your salary won't work anymore. What's already been observed is that people who are just job hoping without any major form of upskilling, without really improving their skills, are roughly getting the same salary as the people that are just staying in a job.

This is not an employee market. This is an employer's market. And unless you up skill, unless you deliver more value, there's no reason for the company to pay you more if they have 50 other people waiting in line. If you want Bogdan to help you up skill yourself, get to senior and get the pay increased next time you switch jobs, then check the free training in the comments. Prediction number 10, web innovation will be mostly driven by AI companies trying to justify their valuations.

We've seen that with Anthropic buying one, but my guess and my speculation is that these companies are trying to give investors the image that they are building something similar to what the internet revolution did decades ago. And in that intent of them trying to add more substance to their valuations, they will try to add as much tech as they can to their companies. Prediction number 11, React becomes the absolute default choice for front end apps, whether we like it or not.

In the last 5 years, React basically exploded, if you're a front end dev you know that. A confirmation for that was actually React moving away from meta and introducing the React Foundation, which is definitely converting React into a standard. When it comes to building UI, it gives me a lot of hope, and for you the front end engineers out there, if you don't know React yet, then learn it, and if you already know it, get better at it because it will be with us years to come.

Prediction number 12, Gen Z will continue not to care about software engineering. It seems like careers in tech are losing appeal for Gen Z according to survey. I don't blame them when they see all this messaging about AI, about tech companies, firing people so easily. I guess they also care about different things like everyone wants to become an influencer these days and get some attention, which, in my opinion, it's very sad.

Because 10, 20 years from now, we'll have a huge shortage of people who are interested in building things, of software engineers, and not just people who are, you know, talking about stuff. Thank you so much and I'll see you in the next one.