Our 2026 Web Dev Predictions

My prediction: jQuery and PHP still will be alive in 2026

Such a balanced approach. Love it, guys.

Nobody talks about the impending costs and enshìtification of AI itself.

We’re in a golden era of AI where everything seems free and available but AI companies are going to need to recoup their investment. Ads will be coming. Paywalled features will be more prevalent. Limits will be reduced in an AI shrinkflation. Suddenly your dependence on these tools will cost you out and those that laid off workers will be calling them back and begging for real people.

Yes, in the future we might all be rolling our own local AI for free but its training will continue to be behind the curve, so people will buy in to the corporate offerings in order to stay ahead.

Let’s see how this plays out. Great video!

wechat is hell from a privacy perspective. One ring

my new years resolution is to no longer start any hobby projects in js/ts, the web is in another 2013 era right now where all the current tech is all but deprecated but new tech still lacks widespread support

I want to go back 10 years ago where AI would only be mentioned a few times a year, now I hear/see it mentioned several dozens of times a day, it feels like it is becoming a filler word.

Guilty of using Google Docs in dark mode right here!

EVERYONE needs Deno!

Unfortunately, I don't think Web Design will come back sorry to say. Companies are lazy and most would rather force their developers to use bootstrap and make everything look the same and familiar than interesting. I am one of very few who enjoys front-end development in the company I work at. Everyone else praises letting AI tools design the UI with bootstrap and touching up any errors.

JavaScript is way more fun when using it for complicated user interaction stuff. Getting stuck making ui primitives like dialogs, tooltips, toasts, custom selects, etc with a bunch of js every time has always sucked.
Also I’ve come to the conclusion that js on the server just kinda sucks. JS has the best chance to make real full stack frameworks that works perfectly with some front end interactivity in one language but they decided it’s more important to sell serverless hosting and limit frameworks to that. At minimum node needs a server v2 which uses web request and response.

I ended up fed up with all the code I need to undo all the layers that frameworks add, just to get better performance. So, I created my own reactive library, XynHTML, that allows full control over the rendering of the DOM tree. One area I’m focusing on is not having all data immutable, since collections (arrays, maps, etc…) have more performance when you don’t copy the structure, but right now those features are still under development. I’m looking at using web components to create fully isolated reactive components and reduce the boilerplate to stand up custom components. I will always have the internal functions as a public API and will keep those documented.

My prediction is that my predictions won't come out in 2026

Happy New Year Scott, Wes, and CJ thanks for great year of content !

WebGPU? Sure…

I use too many apps at work to have them all in one mode, helps my brain to split them to light (info like docs and tickets) and dark (input like editor, messaging)

"enshittification" - you keep using that word; I don't think it means what you think it means. It's not about software performance or security holes or anything like that, it's about software having exploitative features.

So glad I'm not the only person moving back to light mode. It feels so much more open and free

hey thanks for the great content, this is probably my favorite channel that i started following this year, i only wish i found you guys sooner. really appreciate everything you do

If everyone starts using the newest LLM and puts shaders on their website, it will be the new "AI did this" level effort. It's best to master the tools but stray from the masses.

Light mode causes me to strain a lot it's not matter of taste in my case, it makes the experience measurably worse for me unfortunately 😢

7:04 where’s the syntax unrated channel

Dark reader docs all day

I’ve always been a light mode guy. People have always been aholes about it

Let the AI choose the framework? Seems like the developer should choose the framework that they are comfortable with. Am I missing something?

I love the perpetual-learner attitude

❤

RPC
It can't be stated often enough. RPC is a terrible idea, if not treated with all of the scrutiny of compilers and cryptography systems.

Any time you say "trust me bro, I'm going to give your server some code and all you have to do is eval it, and it's totally going to call the right processes with the right stuff"

...the number of steps that you need to take to ensure:

1. that user can not call anything you don't want them to
2. that users can not pass any data that overrides behaviour that you intend for them to do with that function
3. that the syntax for declaring the function to call and data to call it with can't have anything injected into it that is going to affect ... literally any part of the stack / framework / app, by having an overridden behaviour uncaught by other parts (see SQL injection)

We learned this with XML DTDs, and dynamic API generators that promised to take some schema definition and make a full server stack. "If your XML has a DTD that references a class that your system knows about, we can automatically instantiate it and call it, no problem"
...no, that's a huge problem.
"Oh, we could even go further, and with a user-provided schema, we could dynamically generate the class, and let it do what it needs"
Bigger problem.
"Oh, the user could give us an arbitrary URL to a schema, and our framework will download that, and convert it to a class, and then call it the way the RPC intended to call it"
Biggest problem.

There are mitigation strategies. Most "RPC" isn't actually directly RPC, anymore... like, it's not interpreted as running code, directly, and not treated like an arbitrary lookup to literally anything that the framework lets the server see...

...but if I have a mapper in between, that maps the procedure as a string / number / etc, to some closure that calls the right thing:
`{ "getMyUsers": (input) => users.get(input) }`
then there becomes little meaningful difference between it and REST (when not taken as a religion).
The difference between that mapping and
`{ "/x/y": (input) => /* ... */ }`
is virtually nonexistant.

...between NoSQL vulnerabilities, RSC's RPC server actions, Log4J... we're just trying to hit all of the OWASP's Top 10 bucket-list typics.

2026 will be the year Ladybird kills Chrome

I predict I will throw a record amount of javascript kiddie resumes in the trash this year.

Oxlint and Oxfmt both bet on interoperability with the "existing plugin ecosystem": Oxlint with ESLint plugins (which already works well & is fast), and Oxfmt for prettier plugins which is one of the major goals before 1.0 of the formatter.

My prediction for 2026: massive layoffs of vibe coding bros who’s coding skills completely atrophied

I have a feeling the editor will change again for me. *shakes fist at clouds