This Chart Scares Me

Added: 2026-05-30

[00:00:00]The results for the state of webdev AI survey for 2026 just dropped. And this survey is from the same people that bring you like the state of CSS and state of JS survey. So it's really reputable, but it's more geared towards web developers than anything else because that's the audience of people taking this survey. And it also tends to skew a little bit more towards people that are more focused on AI just because it's an AI focused survey. But I went through all the results and there's some really interesting things that I want to show you. First of all, in this use cases section is just talking about what do people use AI for the most? And the really nice thing about this is it'll actually show you this hash section, whether it's a white hash or a gray hash, what is increasing and decreasing between the previous year, 2025, when this survey was taken. So you can see most people are using this for code generation, code review, learning, debugging, summarization. These are the things that AI tends to be the best at.

[00:00:46]So it makes a lot of sense that people are using these tools like this. Now, you will notice there's essentially a massive increase in code review. My main thinking for this is maybe this was a write-in answer on the last test because this seems like a very large increase.

[00:00:58]But code review is definitely something that AI is really quite good at to be able to help you out with. Now, the thing that is the most interesting in my opinion to look at is this AI code generation section. And I actually have a bunch of different graphs that I want to show you. But just first of all, you can see the massive change from 2025 where the majority of people are in the I don't really use that much AI maybe 25% of the time to now the majority of people are in the I use AI 75 or so percent of the time. You can see right here the average sits at 54%, but you can definitely see there's a massive skewing going towards this higher AI percentage. And this 100% AI category has drastically increased from what it was before. But this kind of doesn't quite paint the full picture. One really interesting thing to look at is when you compare this to the experience level of the developer. So this actually will show you the people that are juniors. So 0 to four years of experience. We have 5 to 9 years experience. That's more of a mid-level developer. Then this graph right here is 10 to 20 years of experience. So kind of that senior level developer and then we have 20 plus years of experience here on the bottom right hand corner which is like your most senior developers. And the biggest thing that I noticed when I compare these different charts is this 100% AI category is 17% for junior developers but it's not even close to 10% for any of these other categories. This means that 10% more junior developers are writing with 100% AI versus all the other categories. And you can actually see this average number drops from 63% down to 51% when you're just moving up the experience level. And every time you jump a level, that percentage drops a little bit. So you can see here that the less experience someone has, the more they rely on AI to write the code, which makes a lot of sense because AI is pretty good at writing code. And if you're just getting started, you probably struggle and are slower at writing code. So the AI is like a crutch you can rely on to write relatively good quality code without needing to actually learn what you're doing and writing the code properly. And since more than likely these senior developers are the one doing the code reviews of your code, you don't have to worry much about quality because these senior developers are going to catch that problem for you.

[00:02:49]Which is why I see think we see this trend of a much higher AI usage for the junior developers. Now this direct jumps directly into the next category which is how much do you think using AI forces you to rely upon it. So relying on AI tools will result in less skilled developers overall. And this I broke down by the amount of AI that people use. And the really interesting thing is is the higher you score on this category, the more you think relying on AI will make developers worse over time, while the lower the number, the less you think there is correlation. And as you can see, it's a direct line there. The less you use AI, the more you think using AI will make you a worse developer. While the more you use AI, the more that you think using AI will not affect your ability as a developer.

[00:03:25]And I think this is an interesting thing cuz I would have expected it to be almost the opposite direction cuz I find the more that I use AI in writing my code, the less that I understand the code, the less I understand what I'm doing, and the less critical thinking skills I develop when I'm writing that code, which makes me a worse developer overall. So I find the more AI I use, the worse I get as a developer. But you can see here that chart is not showing that direction at all. Now the next interesting comparison we can make is based on how they are actually using this code for refactoring purposes. So the question asked when using AI to generate code what proportion do you rewrite or refactor before you use it?

[00:03:58]So let's say you get a bunch of code generated by the AI how much do you manually change or refactor yourself?

[00:04:03]And it doesn't even actually say manually rewrite. It just says rewrite or refactor. So essentially are you taking the code that the AI generates 100% or are you changing parts of it that are being generated? And you'll notice here a really interesting trend.

[00:04:14]The more people rely on 100% AI to write their code, the more they just accept whatever the code is without making any changes. You can see here 50 percentile for this category means that they only rewrite 13% of the code being generated by the AI. And this isn't them manually rewriting it. It's just asking the AI to change 13% or so of the code that's generated. That means in essentially 90% of the cases, they're accepting 100% of the AI code exactly as it is the first time without changing or refactoring in any way. While obviously you can see people in the zero AI percent category are much more likely to refactor a large portion of the code being generated by the AI. I think this is a really interesting trend because the more you just trust and rely upon the AI to write your code without refactoring, changing it, or reading it, the more likely you are to end up with sloppy code or bugs that actually sneak in without you even realizing it. Now, this last category that I want to talk about here is going to be based on what kind of code do you generate using AI tools. So, this is asking people when you actually do use AI, what is it being used for in your codebase? And you'll notice here for the lower percentage of AI, by far the number one answer is going to be helper functions and test. And that makes a lot of sense because AI is really good at writing small functions that are like pure functions and they're pretty good at testing out your code. So it makes a lot of sense that these are like the highest categories for people using less than 50% AI. Now, if we swap that and we go to people using more than 50% AI, we see that a lot of them are writing front-end components, which is interesting because I find that AI is definitely pretty bad at writing UI related code, especially CSS. That's one of its weakest points in my opinion. And also, you'll see things like core app logic is rated much higher for these people using more AI than people using less AI. This is actually a big problem in my opinion cuz this core app logic and API integration code, those kinds of things are some of the most important things to get right when you actually write your code cuz they're like your core business logic. And if you're only refactoring or looking at 13% of the code the AI generates, you're going to end up with a lot of potential bugs or errors in that core app logic, which is the most important thing to do right the first time. So, I think it's really interesting to look at what's actually being generated. While on the lower AI side, it's more like helper functions and things the AI is really, really good at writing and even if it gets it wrong, it's not a big deal or it's easy to catch. While people that are using more AI are having it do things that are harder to catch inconsistencies or problems in because they touch more paces in your code. They're much more wide sweeping or they're things like front-end components which are just notoriously more difficult to test. Now, we can look at the general answer to a lot of these questions as well. For example, you can see here a really interesting trend for this is the refactoring section. So when using the AI to generate code, what proportion do you actually end up changing? And you can see that this has dropped significantly. It used to be that people were generating the code and changing a lot of it. But now you can see the vast majority, or I shouldn't say the vast majority, it's about 50/50, but a much larger portion of people are in this category where they're not actually refactoring very much of the code being generated by the AI. Now, this could be one of a few things. This could be the AI getting better, which I will say the AI does get better over time, so you have less need to change certain things cuz it's getting better. or it could be people kind of getting lazier and just relying on the AI to do the right thing and you're forced to generate code so fast because the company you're in, you don't have time to actually refactor that code. So, there's a lot of things that go into this or you're just pushing it off to someone else. You send up a PR and have someone else review that. So, you don't need to refactor the code cuz someone else is going to review it and then refactor it for you. There's a lot of stuff that kind of plays into this, but it's an interesting statistic nonetheless. Now, here I think is honestly the most interesting category is the reason why you end up refactoring the code. And here we have poor code style is massively increased.

[00:07:31]Hallucinations and inaccuracies essentially wrong code that's a massive increase. Poorly readable code which I think falls in poor code style. So just bad quality code that's wrong. Excessive repetition again lowquality code faulty code again bad code. And then when it comes down to things that are easier to fix like variable renaming, outdated inputs and things seems like the AI has gotten better at those particular things. But when it comes to the important stuff like writing highquality correct code, the AI is really really still quite bad at that. The interesting thing I think is that a lot of people are pushing like we should write with AI. It's so fast. It's so amazing. But would you hire someone just because they're fast? Imagine you go into a job interview and you say, you know what, I can write code 10 times faster than your entire development team, but I'm going to lie to you. I'm going to write bad code. It's going to be really difficult to maintain and it's just overall going to have problems inside of it. Oh, and also I'm going to cost way more than any other developer out there. like would you actually hire that person? Probably not. But people still think that AI is the solution to this. Yes, I think that the faster that you can write code, the less that code quality matters to a degree because code quality being bad means it takes more time to write code because you now need to fix bugs and you have more places to change your code.

[00:08:38]But if you can write code at a certain speed, you can kind of get around bad quality code by just brute forcing speed at it and writing your code faster and faster. So even if it takes you 10 times as long to actually implement a feature, if you're 10 times as fast, it evens out to the same as you were before. I think where we're going to start to see this whole idea fall apart with code quality is as you start to scale to larger scale applications or as we start using AI for years and years on the same codebase that keeps getting worse and worse.

[00:09:03]Eventually the code case is going to get so bad that the amount of time it takes to add a new feature into that codebase is so long that if you would actually to spend the time to write good quality code to start with, you would able to be iterating faster than this AI that is 10 20 times faster than you as a human. So I see that as a problem that is going to be happening going forward and is honestly already happening at a lot of companies and it's not surprising that the biggest problems with AI is that it writes bad quality code and incorrect code. Some other interesting statistics down below is how often do you generate code? So this is just how frequently do you use AI. You can see here people are using it more than they were before.

[00:09:35]That makes a lot of sense based on the previous graphs and you can see that it's definitely showing that a lot of people are using it constantly compared to what it used to be. Now if we look down here, this is just people using AI for things other than coding. Again, this has increased as well, which again makes a lot of sense based on what we've seen. But also, this never category has increased as well, which I find a little bit interesting. It's not a massive increase, but it is an increase nonetheless. And I think that could just be people realizing that they do more with code so they don't need to use AI for other things or they find that they just have problems with AI doing other things and they spend more of their time on coding. Not sure why this is but it is an interesting statistic nonetheless.

[00:10:08]Now if we look here at generated code we can see what people are increasing the frequency of. So we can see helper functions has actually decreased which I find hilarious cuz that's one of the best things that AI is good at. And you can see front-end components has increased which again is something I find that AI is worse at than almost anything else. Testing has increased which is great. documentation and comments. Another thing that AI is good at. And now we come to core app logic, API integration code. Those are definitely things that the AI struggles a bit more with and you should definitely babysit. And if we go down here, CSS code, I have not seen very good CSS code from AI. So, this is interesting that it's increased so much because it's something that I find that AI is really quite bad at. We also have a massive increase in database queries.

[00:10:44]This kind of makes a lot of sense cuz database queries are one of those things that are really a pain to write, but once you actually get it written, it's not too hard to actually manually test and make sure it's correct and make sure it's optimized. So this is actually a place where I think AI is quite good at writing database queries and then you can go through and test and make sure to optimize them yourself. But the actual writing of the raw query can be a little bit of a pain depending on the level of your SQL skills. Now another interesting category is how much do you personally spend on AI per month? And you can see the big increase is in this $100 to $500 a month tier. And that is mind-blowing to me. That's over $1,000 a year. And if you're spending, you know, $500 a month, that's way over $1,000 a year. You're looking at $5,000 a year just on personal AI usage. That is a crazy expense in my opinion for a lot of people. I mean, I don't pay $100 a month for any AI tiers and I think spending that much money, yes, you can get your money's worth out of it, but these are only increasing in cost. You can see the amount of people spending 0% drastically inc decreased. And you would think maybe this 1 to20 tier, this $ 20 to $50 would see a lot of the gains from this decrease right here, but instead they're jumping all the way up to this like $100 plus p plan. So AI is really pricing out a lot of people that can't afford to pay 100, 200, 300, $400 a month. And as we've seen with AI pricing, it's only getting more expensive. The really free, you know, cheap tiers we're getting, those are being removed and replaced with usage based or much higher tiers or just massive throttling of your limits.

[00:12:00]And I see as the years go on, this price you need to pay to get the same amount of AI is going to just keep increasing to higher and higher categories to the point where it's not even profitable for you to use AI on your own. Now, kind of the last question that I want to look at in this section is going to be local AI.

[00:12:13]And that's essentially what amount of people are interested or using local AI?

[00:12:16]And you can see here the yes category has increased slightly probably because of the prices of AI coming up so much.

[00:12:21]And if you want to actually learn how to set up your own local AI models 100% for free on your computer, I'm going to have a video which is linked on your screen right over there. It's a massive crash course showing you how to set up everything you need to know about how to run AI models locally on your computer that can give you some really quite incredible speeds and some really decent results as