My 2026 Predictions

00:00:00: 2026 will for sure be another interesting year and I think

00:00:04: in some regards, like the job market, it will be better than

00:00:08: 2025 for developers. So I'll share my perspective

00:00:11: on AI, obviously, the job market and why I think

00:00:15: it will get better, and web dev and web dev

00:00:19: trends and technologies in general.

00:00:22: So let's start with AI because that obviously will

00:00:26: still be a very important thing next year, and here

00:00:30: I think we have kind of reached a plateau when it comes to the

00:00:33: intelligence, if you want to call it like this,

00:00:36: does not mean that we won't see more development

00:00:40: or better models or tools around these models

00:00:44: next year, and let me explain. So if we had a chart and we would

00:00:48: start in 2022, obviously where ChatGPT was released, and

00:00:52: then we would end at the end of 2025, I would say the

00:00:56: "intelligence," and I put that in quotes because they're not really

00:00:59: intelligent, they're token generators, but of course they do that pretty

00:01:03: well and they can use tools and I'll get back to that.

00:01:06: But the intelligence, the capabilities of these tools,

00:01:10: I would say developed something like this maybe, so a

00:01:13: pretty good development. Obviously we all saw that, but I

00:01:17: think that now we're kind of at this point, so I don't think

00:01:21: we'll see an exponential improvement or

00:01:25: anything like that. Instead, my feeling for the last year whenever

00:01:29: new models were released was that there was a lot of hype, there

00:01:32: were benchmark numbers where the model was better than most other

00:01:36: models of course, but at the same time not that much

00:01:40: better. And especially once people started using

00:01:44: than once that it was kind of meh in the end.

00:01:47: I mean, take for example the LLaMA 4

00:01:51: last year. When LLaMA 4 was announced, the model

00:01:55: family, we had impressive benchmark numbers.

00:01:59: We saw that the top LLaMA 4 model, maverick,

00:02:02: basically was better than all the other top models

00:02:06: or other frontier models we had at that time

00:02:10: in benchmarks. But shortly after the release,

00:02:14: this where people were kind of, "It doesn't feel that

00:02:18: great," and we had that more than once last year.

00:02:21: Now don't get me wrong, models did get better for sure,

00:02:24: especially of course compared to 2022

00:02:28: or 2023 or 2024 and so

00:02:32: on, but we're at this point where they won't

00:02:36: get drastically better every year, and I'm talking about

00:02:39: large language models here. If we take a look at video generation

00:02:43: Veo or Sora 2, obviously we saw a

00:02:47: drastic improvement there compared to the years before, and the same is true for

00:02:51: image generation models. We had the release of Nano

00:02:55: Banana, not the pro version but the normal version, in

00:02:59: August of this year I think, and that was really groundbreaking.

00:03:02: And then of course we also had Nano Banana Pro which

00:03:06: raised the bar even more, and we had other models like the

00:03:10: Flux models by Black Forest Labs, Flux 2 recently

00:03:14: released, and all these models definitely got much

00:03:18: year before. But if we take a look at large language

00:03:22: think we're reaching that plateau.

00:03:24: Now I will say there's a lot of development on

00:03:28: those models to make them better at specific tasks.

00:03:31: I think that's the clear trend we can already see that will continue in

00:03:35: 2026 because of course we now have those thinking

00:03:39: models. That is pretty much the standard now for

00:03:43: models. We also have models that are better at following

00:03:46: instructions and at

00:03:50: tool calling, and that is a big one.

00:03:53: This here really is a big one because that is

00:03:56: important. It's that improved tool calling

00:04:00: capability of new models that makes them way more

00:04:04: useful in the applications that use these models,

00:04:07: and that is important. With this chart here,

00:04:11: I'm referring to how models or their raw

00:04:15: intelligence kind of improves. Now if I were to

00:04:19: copy this chart and judge the capabilities of

00:04:23: applications that are built on top of these models, it would

00:04:27: probably look more something like this, more linear, not

00:04:30: exponential still but linear. Because when we look at applications that

00:04:34: use AI like ChatGPT or Gemini themselves

00:04:37: obviously, then they got better. Oh, also of

00:04:41: course because all these models have great

00:04:46: multimodal capabilities. They're good at, uh, taking a look

00:04:50: at images, understanding images, generating images by

00:04:53: integrating other models, and that's basically what these

00:04:57: applications are doing then which takes me back to this chart

00:05:00: here. It's important to understand that of course we

00:05:04: are exclusively almost interacting with applications

00:05:08: that use AI, unless you use the raw models via the API

00:05:12: to build your own applications of course.

00:05:13: But as end users, we interact with applications like

00:05:17: ChatGPT or Gemini of course,

00:05:21: so that would be example applications we have, but

00:05:25: also something like Cursor or

00:05:28: Claude Code. These are applications built on top

00:05:32: of AI models and those of course got much better over the

00:05:36: year and I think they will continue to get better. Why?

00:05:40: Because the developers of these applications get better at

00:05:44: integrating those AI models, there's a lot of wiring of different

00:05:47: models behind the scenes, there is of course prompt and context

00:05:51: engineering, and there are tools that are exposed to those

00:05:55: models. If we take a look at ChatGPT

00:05:59: here, the ChatGPT of today is of course nothing related to the

00:06:02: ChatGPT we had in 2022. It has, for

00:06:06: example, web search. It can,

00:06:09: uh...... write and execute code. It can

00:06:14: generate images and that, of course, is technically done through a

00:06:17: different model, not through the GPT large language

00:06:21: model, but through an image generation model,

00:06:25: ChatGPT application that wires these different models

00:06:29: together. And other applications like Claude Code

00:06:33: are also potentially wiring different models together, doing

00:06:37: prompt engineering, and they expose tools to these models

00:06:40: like web search, which is exposed by ChatGPT and also

00:06:44: by Cursor. It can also search the web.

00:06:47: And that is one part where those models, of

00:06:50: course, did get better because they are optimized and

00:06:54: fine-tuned for being good at tool calling.

00:06:58: Now, just to make sure we're all on the same

00:07:02: a tool really just means that the developers of an

00:07:05: application describe the available tools to the

00:07:09: model and then the model emits

00:07:12: tokens that describe an intended tool use

00:07:16: if it feels like calling that tool.

00:07:19: Now, of course, whether it feels like calling a tool

00:07:23: simply on the prompt engineering, the system prompt and how the model

00:07:27: has been trained and fine-tuned. So for example, if a tool

00:07:31: has a get weather tool available and the user asks it

00:07:35: what will be the weather tomorrow, there is a very high

00:07:38: likelihood for the model to suggest that it wants to call the get

00:07:42: weather tool. Now as a next step, it will emit those tokens

00:07:46: that describe that tool call and then it's again the

00:07:49: application like ChatGPT that will make

00:07:53: the actual tool call. So that it will execute the actual code

00:07:57: behind the tool and that will then pass the results of

00:08:01: back into the chat history, so to say,

00:08:05: to have the large language model then generate the final

00:08:09: output that should be shown to the user.

00:08:11: So all that tool calling and these intermediate results are not

00:08:15: necessarily shown to the user, but the final result, which

00:08:18: call, is. And that, of course, is where those models did get

00:08:22: much better, or the applications I should say,

00:08:26: they will continue to get better because we, as developers, can leverage

00:08:30: those large language models that have been optimized for being good at

00:08:34: tool calling and so on to build more capable applications on top of

00:08:38: them. So that is the trend I see continuing

00:08:42: I see Cursor or Claude Code and all those other

00:08:46: applications and all those applications that are not related to coding at all

00:08:49: become even better in the next year, not exponentially,

00:08:53: but linearly I would say. Now what does that mean for

00:08:57: us developers? Well, it means two main things I would say.

00:09:01: For one, it gives us opportunities.

00:09:04: There are opportunities for building interesting stuff

00:09:08: on top of AI, on top of large language models,

00:09:11: models, and potentially combined capabilities.

00:09:15: Like the Build My Graphic app I built, I built it for me because I

00:09:18: need infographics for videos like this and it's not just about infographics.

00:09:22: You can generate all kinds of graphics there.

00:09:25: And I built this on top of Nano Banana Pro, but also

00:09:29: by using many other models to process the user input,

00:09:33: do research, whatever, and then generate graphics based on

00:09:37: style templates provided by me. So I built an application on top of

00:09:40: think there will be many opportunities for small and bigger useful

00:09:44: applications that you can build on top of AI because the

00:09:48: AI models have all these capabilities now which they didn't have two

00:09:52: or three years ago. So I think the first important implication

00:09:56: takeaway is that this gives us opportunities.

00:09:59: The second important implication for us developers, I

00:10:03: think, is that we must be able to

00:10:07: work with AI assistants

00:10:10: for coding. And I know that's not everybody's cup

00:10:14: of tea, coffee, whatever. You, of course, don't have to

00:10:19: if you have the luxury of not needing to, but if you

00:10:22: are in a job and want to keep that job or want to

00:10:26: find a job, you really need to build the skills to

00:10:30: effectively work with AI assistants for coding.

00:10:33: And you might wonder, what does that mean?

00:10:36: I mean, I can write a prompt. It's about writing the right

00:10:39: prompts, asking for the right things, providing the right context, not

00:10:43: overwhelming the model and asking for too much,

00:10:47: small, incremental demands or requests.

00:10:51: And I will create separate resources on that to share more about how I

00:10:55: work with AI and how I think you can effectively work with AI

00:10:59: because I definitely want to help there.

00:11:01: But I think you need to build that

00:11:05: skill unless you're really in a position where you're your own boss and you can do

00:11:08: whatever you want, and then of course it's totally fine

00:11:12: AI at all. Let me be very clear here.

00:11:15: I'm also not saying, and I want to be clear about that too,

00:11:19: that thing that will generate all perfect code and that we should

00:11:23: go into live coding. Quite the opposite.

00:11:26: I'm talking about AI assistants here, not live

00:11:29: coding. You as a developer can and should

00:11:33: steer and control the model. You should tell it what

00:11:36: to do and you should get good at being precise there about

00:11:40: providing the right context, about being able to evaluate the results

00:11:44: you get back, about building a plan before you even

00:11:48: prompting. And again, I'll share more about that.

00:11:50: But that is the, the second important imple- implication, I

00:11:54: think. That is my take on AI

00:11:58: in this, uh, new year, uh, from a developer's perspective.

00:12:02: Now of course, AI is one thing,

00:12:05: arguably the most important one, but let's talk about the

00:12:09: job market because that clearly is also important to many people.

00:12:13: And here, we probably all know this chart here, this

00:12:17: horrible chart where we had this peak here during the

00:12:21: pandemic and, and now we're down here.

00:12:24: Now the good thing, I guess, about this chart is that it looks like we,

00:12:28: bottom, right? Maybe there even is a slight improvement coming for us

00:12:32: or already happening, and I definitely think there will be an

00:12:35: improvement n- next year. Now we also have to differentiate

00:12:39: of course. We have to differentiate between jobs for

00:12:42: senior developers, and I'm only talking about the developer

00:12:46: here, and junior developers. I think for

00:12:49: senior developers it was already okay or good

00:12:53: and it definitely will be good in the next year because all those

00:12:57: companies are scrambling to get senior

00:13:01: developers, especially the ones that are capable of working with

00:13:05: AI, of steering the AI, and of judging

00:13:08: the output by AI. So senior developers

00:13:13: h- have a relatively good position I'd say,

00:13:17: get much better in the coming year.

00:13:20: The problem, of course, is that not everybody's a senior

00:13:24: become a senior by being a junior first and

00:13:28: here, we definitely had a- a bad (laughs) state

00:13:32: of that job market in 2025. But the good thing is,

00:13:36: I expect that to get better in-- not

00:13:39: badder, (laughs) but better in the next year.

00:13:43: In 2026, I expect that to

00:13:46: improve. Now, why do I expect that to improve?

00:13:49: Of course, we have all these reports about layoffs.

00:13:51: We have all these reports about layoffs because of

00:13:55: AI that happened in, uh, 2025 but we

00:13:59: have to be careful here. It's pretty easy

00:14:03: to mention AI as a reason for a

00:14:06: layoff. That's a nice reason to have.

00:14:09: It's especially a reason that might please investors because,

00:14:12: they wanna hear that you're an AI first company and that you're

00:14:16: embracing AI. And I'm not saying that AI is not

00:14:20: having any impact on the job market, it clearly has, and even if it's

00:14:24: just that many executives expect AI to be able to

00:14:28: do your job in the future, so they rather don't hire anybody new for

00:14:31: now. I'm pretty sure though, that more and more companies will

00:14:35: realize that going all in on AI

00:14:39: or relying heavily on AI instead of humans that made and

00:14:43: work with AI, is not a good strategy.

00:14:46: I'm sure we'll see in 2026 that more companies

00:14:50: want to have a workforce that's capable of using AI as a

00:14:54: tool because it is a great tool, at least in certain industries.

00:14:58: Encoding, I would say, it can be a very useful tool but it's

00:15:02: that, a tool, not a replacement, and I think more

00:15:05: companies will realize that in 2026 and they will

00:15:09: also realize that yes, of course, it would be nice to have

00:15:13: more seniors that are capable with

00:15:16: AI, that's why this will be such a good job

00:15:20: think, but that there aren't enough seniors

00:15:25: squeeze and that's why I think more companies will be open to hiring

00:15:28: juniors again. But the role of

00:15:32: developers will change, not just next year, but in

00:15:35: general. As I mentioned, it will be more

00:15:39: about working with AI, about controlling AI,

00:15:43: and junior developers that feel confident working with AI

00:15:47: will be more interested next year than junior developers or developers

00:15:51: in general that don't. And therefore, that is one

00:15:54: important takeaway for next year, the first important

00:15:58: takeaway. As a developer that wants to get a

00:16:02: job, you must be able to

00:16:06: use AI. And again, sounds like a simple thing but

00:16:10: you must feel confident using it. You must be able to show potential

00:16:13: employers that you can get a real value out of

00:16:17: AI assistance. That's one thing, but I also

00:16:21: think there is a second thing, unfortunately, that will be important

00:16:25: as a junior developer, especially if you wanna get a job next

00:16:29: year, and that second thing is that you will need connections

00:16:32: and/or a strong social media presence or

00:16:36: anything like that in addition to having a good portfolio, a good

00:16:40: resume, and so on because there is this huge crowd of junior

00:16:44: developers. I mean, look at this chart, right? We're coming from here.

00:16:48: Lots of people worked in the development industry, lots of people

00:16:51: started getting into it because there was all this

00:16:55: demand here and therefore, all these people are

00:16:59: still here today but there are less jobs.

00:17:02: So in order to stand out from that crowd

00:17:06: here, you will need something that makes you stand out

00:17:10: and for sure, that is proving that you can work with AI,

00:17:13: employers will want that, but it will also be

00:17:17: about, well, having connections, having a strong social

00:17:21: media presence, anything that makes you stand out.

00:17:23: And I know that these are not great news because you wanna work

00:17:27: not as an influencer. I totally get that.

00:17:30: I just think that will be something that will be needed to

00:17:33: vastly improve your chances of getting

00:17:37: that job. And again, you still need that

00:17:41: portfolio and all that fun stuff, but that will be important

00:17:45: too. So that is the job market next

00:17:49: year. The last thing I wanna take a look at

00:17:53: dev in general, so the fun part now. Which trends do I see there?

00:17:56: Which technologies, uh, should you learn and use?

00:18:00: And here, I would say we have one

00:18:03: clear default stack, the standard

00:18:06: stack you will use in most projects next year and that

00:18:10: is TypeScript, React,

00:18:14: Next.js, Tailwind. That is

00:18:17: AI's favorite right here. That is what you get in

00:18:21: most, uh, applications if you just ask AI to generate something

00:18:25: for you, some web app. Uh, that will be important next year too.

00:18:29: Now, by the way, I of course have the JavaScript angle

00:18:33: PHP world. Let me be very clear, PHP, Laravel, all

00:18:37: amazing technologies which can land you jobs.

00:18:40: I just have that JavaScript lens, which is why I'm talking about that.

00:18:43: But if we take a look at the

00:18:46: ReactJS download numbers, for example,

00:18:50: it's pretty clear to see that AI has a

00:18:54: favorite here in web development in general and especially the

00:18:57: JavaScript web development world, of course.

00:19:00: So that is that default stack we'll have next year.

00:19:03: That is the stack which you should learn

00:19:07: are that most companies or many companies will be using

00:19:11: that. Now-If you have the luxury

00:19:15: of deviating from that because you're not looking for a job

00:19:18: your own, or because you wanna kind of expand your

00:19:22: wanna have more than just one stack that you bring to the table,

00:19:26: I think for next year, one interesting thing to dive into would

00:19:30: an alternative to Next.js like React Router or TanStack

00:19:34: Start, which is amazing and TanStack Start,

00:19:37: especially something I wanna cover more next year.

00:19:41: I think it also might be interesting to look at modern browser APIs

00:19:46: like the, uh, Popover API, for example, which I recently

00:19:50: covered on my other, uh, channel and I'll do more about

00:19:54: and about CSS because a lot

00:19:58: has improved here. Uh, raw

00:20:01: CSS and also what the browser can do for you has gotten

00:20:05: so much better in the last two years.

00:20:08: It's just that nobody noticed because of, well, AI.

00:20:12: It's the big thing which takes all the attention but there

00:20:16: has been such amazing development here that you can get a

00:20:20: lot done by just using what's built into the browser which is available

00:20:24: across all major browsers. So it's production ready and/or

00:20:28: by using modern CSS features like native, uh,

00:20:32: CSS nesting where you can write CSS code

00:20:36: like this and it just works, no compilation step needed, by

00:20:40: using CSS layers to control precedence

00:20:43: and much, much more. There's a lot of

00:20:47: amazing stuff available in the browser right now.

00:20:50: It's also something I wanna teach more in the next year

00:20:53: to, well, bring some more awareness to that

00:20:57: because you might not need all those extra libraries you're

00:21:01: pulling into your project or all that code that's being written

00:21:05: to be wr- you don't need to manage your own modals

00:21:08: code and your own divs which are then maybe not accessible

00:21:13: or not always on top. You can let the browser do a lot of that

00:21:16: work and the result will be better.

00:21:19: And then, of course, I mentioned it before, you can build stuff

00:21:23: with and on top of AI, on top of large

00:21:26: language models, image generation models and so on,

00:21:30: There are all these opportunities, uh, which you as a

00:21:34: developer can try to take advantage of

00:21:38: because you know how to write code.

00:21:40: You can use AI to accelerate the development

00:21:43: process and, therefore, you have these opportunities

00:21:47: stuff on top of AI by using AI, that can really also

00:21:51: be an interesting area next year, and I got exciting stuff to share next

00:21:55: exciting content I'm working on and, um, yeah, I'm,

00:21:59: therefore, really looking forward to the next year.

00:22:01: I think it will be a better year there for developers than 2025

00:22:05: was to some extent. Definitely still challenging,

00:22:09: definitely also exciting, I think.