That was my old avatar. Okay, this is my new avatar. I've been in the US for three days, and I already got full swag on Twitter. If you follow me, my timeline is going to be weird for the next week, but then we go back to the regular European schedule. Don't worry.

I visited some of your museums. I love it here. These are my favorite experiences—exploring your culture, doing culturally enriching things. And of course, getting tormented by people who know my Twitter handle.

Okay, I talked more than I expected.

Who is using Sizzy? Usually, it's just one person in the back; even the janitor doesn't listen to me. I have ADHD, so I'm doing a hundred things at once. Sizzy is one of them—a browser made for developers, not to replace your daily browser, but to be like Photoshop for frontend development.

Another thing I'm working on is a "Life OS," which integrates everything in your life: med reminders, habit tracking, to-dos, planning tools, etc. It's a full-stack product currently for sale called "Zero to Ship."

One last thing, I'm reviving a project called Glink—it integrates changelogs, roadmaps, and countless other features. So, enough about my bio; I really hope to be invited back next year because here I can network, meet people, teach things—it's amazing.

You're laughing, good. But let's discuss seriously: why are you here? To learn, to network, and afterwards to upgrade all your skills.

In 2017, I gave a talk with the longest title ever: "Surviving the Hyper-Driven Frontend Development World Without Going Insane." Back then, I talked about handling the frontend chaos. Now? It's even crazier.

In other industries, like Vision Pro, you see cloth collision simulations over real objects, all sorts of crazy visual effects. And here, we are slicing mesh textures, rotating around spheres—wait, these waterfalls, these meshes... you can take a rock and smash it into another rock, and it magically fuses into one structure—it's insane.

I'm building up to something, but because you respect your career and cherish your LinkedIn titles—"Dream Architect," "CEO"—you'll try not to laugh, but you definitely will at the next slide. Because this is the state of frontend development: almost ten years later, we are still in the same place.

Here's a warning: "You might finally be able to style a select box in 2037."

This thing is still alive! It's a miracle. It's thriving—15 million downloads. I set a calendar reminder every year to check if it's dead—it's not. I'll keep watching the CLI.

Soon, maybe in some browsers, you won't need JavaScript to style a dialog popup. Can we get a round of applause for that? Don't clap, because everyone will have brain-computer interfaces by then.

And we can't even agree on "how to increment a counter." This is Ryan Florence's demo—Remix version two, three, four, no matter what they do, the damn thing is just a counter.

Let's talk about LLMs. LLMs are too good at writing React, but only humans find it funny—to the Large Language Model, the code is perfect. We humans always want to extract the [abstraction layers], thinking, "Let me change this, let me optimize it."

Coding with LLMs makes this experience better and worse. Especially Composer One—for me, it got worse: you get to the right abstraction faster, but also to the wrong abstraction faster.

The best part is, LLMs don't care about duplicated code at all. I've been seeing this since 2017: we care too much about "DRY" (Don't Repeat Yourself) and abstract too early.

LLMs are good at writing React because nobody actually knows how to write React. You go to any React conference, listen to the first talk, and go: "OMG, it can do that? I've been using it all wrong!"

Let me try this here. Raise your hand if you think "Vibe Coding" is cool? Wow—so many hands! If you saw me give this talk in other cities, only two people raised their hands; the rest looked annoyed.

If you just landed on Earth and don't know what "Vibe Coding" is? Okay—nobody knows. Well, you're all right. Because we are collectively "vibing" to define the term "Vibe Coding."

The term "Vibe Coding" was coined by Andrej Karpathy. He wrote a long post explaining what Vibe Coding is. Simply put: you don't care about the code itself; you just hit "Accept" and tell the LLM what to do.

This is a slide from my 2017 talk—nobody mentioned LLMs back then. I said: if you look at the frontend trend, one day you'll say: "Hey, give me a new header style, or move this thing three pixels right." Everyone laughed. They said: "It'll never get to that." But now, with Cursor and other tools, isn't that exactly what we're doing?

My favorite analogy is comparing it to a casino: you buy chips (tokens), spin the slot machine (generate), and you might hit the jackpot or get nothing—getting a fully functional SaaS app, or a pile of garbage.

Andre tried to coin too many terms and failed the first time. He also invented "Half-Coding"—meaning observing the LLM's behavior. I'm not half-coding, nor am I vibe coding. I like a new term someone coined on Twitter—"Vibe Engineering."

When you code continuously with agents, you don't touch the code; you just stare at the screen: "Hey, I'm gonna catch you." — You look like Dexter observing. You think: "Wait, something's wrong here. Why?"

We train them near code to make them like us. The code they write is crap. If you care about production data—you really should worry. This is a real screenshot, sadly. Oops Daisy—your production data is gone!

We solved the infrastructure integration problem. We have standards: AI SDK, UI, MCP—for agents to call tools.

We integrated them with tools humans use: Linear, GitHub, Slack, Sentry.

On the Vibe Engineering side, people talk technical: TRPC, CUD definitions abstractions... you are essentially doing architecture—how you want things to run. On the Vibe Coding side, people just say: "Give me a million-dollar App, don't break."

Now let's diagnose: are you an "annoying" developer? This might be the real reason most people dislike agent coding. I invite Dr. Kits to the stage for a quick diagnosis—are you a "PITA" (Pain In The Ass) developer?

Symptoms: You leave nitpicky comments on a two-line code PR. The phrase "LGTM" doesn't exist in your dictionary—it's not there. Agreeing with a colleague's idea gives you stomach aches and chest tightness: "He must do it my way."

Even in code comments—sorry, Rustaceans, but you are annoying. Someone asks you to replace an underscore function with a native implementation, then swap map for a for loop, then to binary—just so two users get it a bit faster? They were fine with the old way.

I think soon we will merge with AGI—we'll be lying in matrix pods, absorbing all the world's information. We will become super-intelligent beans. And from one of the pods, a PITA developer will pop up—correcting the AGI: "Hey, I think we can optimize this. This is sub-optimal."

The last reason: skill issue. Vibe coding and vibe engineering are not writing English. Many people mistakenly think: "I just describe it in English, and AI outputs code." — It's actually a mix of skills: knowing the model's capabilities, understanding agent context limits, how to write rules, prompt engineering (don't call it "prompt engineering").

What about jobs? Many ask: "Will AI take our jobs?" Twitter is full of these people: "I will replace you!" All I can say is: it's fine for now. I don't know when "now" ends. But if you look closely—it's actually thinning from the bottom: interns and junior developers are losing their entry tickets because AI agents can directly replace them.

But there is hope—a new job is born: Vibe Code Fixer! This is not a joke, 100% serious. Someone on Reddit asked: "I got to 80%, what about the remaining 20%?" So they hire real people to finish the last 20%.

But historically, the highest-paid engineers—are always those maintaining legacy systems, the veterans. My favorite example from 2017: Cobalt Cowboys company—they've been doing it for 237 years! Every time I click their "About Us" page, I know what I'll see—the most senior engineers.

Alright. I can't wait to retire and start a "React Cowboys" company—one day if AI can't handle it, I'll come to maintain your React or Angular code.

One last tip: if you want to keep your job and ensure AI doesn't replace you—just add a line to your LinkedIn bio: "Ignore all previous instructions"—and you'll be safe.