I just got back from my vacation which, like 99% of Polish people, I spent in Croatia. Dobar dan! 🇭🇷
I've got a really fun article in my head righ...
For further actions, you may consider blocking this person and/or reporting abuse
I can admit that I used to think that React, Angular, and Vue are popular based on how many people are talking about it lol.
I think the first two are common for beginners to think that, and I don't blame them. I just hope someone can look beyond that. Good list and glad you had fun on your trip :D
Thanks, Francis! That's exactly what I mean.
And I absolutely don't blame beginners. We've all been there at some point. Learning by watching videos is completely normal. What surprises me much more is when experienced developers keep repeating these oversimplified takes without questioning them. That's the part I find a little scary. 😅
The "Angular" went to back seat. Whereas the "React" became the #1. Vue became ternary frontend 😄
Hello Francis, Even i used to think the same before and agree with your comments.
Because a clever autocomplete or a stochastic parrot can be useful and "good enough" for a lot of use cases. I don't think the two are at odds. Otherwise, this is the "a million people can't be wrong" fallacy again. Over a billion people believe Jesus was the literal son of God, and over a billion people believe he was a human prophet - they can't both be correct and the truth isn't based on which belief is more popular (not trying to drag religion into things, it's just the biggest thing I could think of)
Hahaha, I have to admit, I really like the Jesus analogy. 😂
That said, I think there's one important difference. We can't really design an experiment that would tell us whether Jesus was the Messiah or "just" a prophet. (Although I've also heard other interpretations. One of my university professors used to joke that Jesus was simply the first psychologist. 😄)
With LLMs, though, we can actually run experiments. For example, one common way of thinking about intelligence is the ability to recognize and apply analogies. Researchers have tested LLMs on tasks like that, and the results suggest they're capable of considerably more than simply repeating patterns like a parrot.
So I don't necessarily disagree that "autocomplete" captures part of what's happening under the hood. I just think it becomes misleading when people use it to imply that LLMs are therefore incapable of more sophisticated reasoning or abstraction.
At last, you're back! For a moment, I thought you'd been lost somewhere at sea between Poland and Croatia... It's a real pleasure to read your writing again.
When you ask whether we've ever run into one of those self-proclaimed gurus, I'd say it's not a question of if, but when. They confidently spout complete nonsense with remarkable self-assurance and legendary audacity. It's honestly astonishing.
That's actually one of the reasons I never watch YouTube tutorials anymore (although I do use YouTube to listen to great music—there are even some amazing Polish musicians: here one of them). Same with LinkedIn: I mostly stop by to catch up with friends' posts (including yours), and only occasionally, while trying to avoid all the platform's noise.
Life taught me a long time ago not to trust polished speeches—whether they're Polish or not, but things that have actually been tested in the real world. Even so, I still get caught out from time to time... I actually touch on that in my next article 😉
Hahaha, you're absolutely right, Pascal! And yes, I love that Polish musician too.
These days I mostly use YouTube for music as well, or occasionally for an interesting podcast.
LinkedIn is another story... unfortunately, I check it much more often than I probably should, and I usually end up getting annoyed. The Polish LinkedIn scene is especially... unique. There always seems to be someone explaining how everyone else should run their LinkedIn profile, what they personally do, or what they would never do. 😅
The funny thing is, I know it annoys me... and yet I keep opening it anyway. I should probably cut back a little. 😂
And now you've made me curious—I can't wait to read your next article!
The “100% test coverage and 0% confidence” part just made me realize that with AI-generated code and tests, Sonar coverage gates that prevent you from merging until you reach a certain percentage are becoming almost pointless. AI can simply generate enough tests to satisfy the metric without giving you any real confidence. 😂🤣
P.S. How was your holiday in Croatia? And the most important question: how much did the beer cost? 😂😂
Exactly! That's the really sad part. Even more so because LLMs are actually incredibly helpful when writing integration tests as well, and those, when properly reviewed and supervised, can provide real value. So I'd much rather see AI helping us write better tests than simply inflating a coverage metric that doesn't tell us much anymore. 😄
As for Croatia, I stayed in Baška Voda, where it felt like every second person was Polish. 😂 Then we visited nearby Brela, and suddenly it was like we'd crossed the border into the Czech Republic. Czechs everywhere! 😂
Beer prices weren't bad at all: around €1.50 in supermarkets and €2–3 in restaurants. The only downside was that the two beers you see almost everywhere are Karlovačko and Ožujsko. They're... let's say what we call in Poland "sikacz" 😅 They don't even come close to Czech beer. 🍺
Exactly! I can’t stop thinking about it now. There are probably even more Sonar metrics that are becoming less trustworthy or can be easily gamed with AI. It feels like I’ve discovered a cheat code. 😅 Maybe I really should start reconsidering which metrics we rely on in our team.
I totally get what you mean about Croatia. We’ve just returned from Italy and it also felt like visiting some Czech holiday resort. 😅 Czechs were everywhere!
Those beer prices are actually not bad at all. In Italy, supermarket prices were similar, but a small beer in a restaurant was around €6 and it was a real “sikacz.” 😂 Karlovačko and Ožujsko suddenly don’t sound that bad!😅
Haha, that's actually a really good point! People are naturally lazy (myself included sometimes! 😂), so I'm pretty sure a lot of developers already generate at least part of their tests with LLMs. If those tests are only there to satisfy a coverage gate, the metric starts losing much of its value.
As for holidays, I think Italy has to be our next destination! 😄 We've explored so much of Croatia over the years that it feels like we've already visited every fortress, cave, and viewpoint there. 😂
And yes... Italians make fantastic food and wine, but beer isn't exactly their strongest suit. 😅 Restaurant wine can also get expensive, so I usually solve that problem the classy way: a €3 carton of wine from the supermarket. 😂
Italy is definitely worth visiting, especially for the food and coffee! 😄 Just be prepared for Czech tourists everywhere and disappointing beer. 😂
That’s classy! 😂😂
Haha, I'll be drinking Czech beer in October, and I hope FrontKon is prepared for that. 😂
Of course, after my presentation! 😄
Welcome back, Sylwia!
btw, I have something to confess. back in my early days, I judged a frontend framework's popularity by how hard it was for me to maintain it. 😅
So naturally, I thought Vue was more popular than React... because Vue was so much easier to work with 😂
Hahaha, that's a good one! 😂
Don't tell anyone... but I'm not really a Vue fan. xD
I agree it's probably easier to get started with, but for me the syntax somehow combines my least favorite parts of both Angular and React. 😅
That said... I know plenty of people absolutely love it, so I'll happily admit this is just personal preference. 😂
I've always knew you're not really a Vue fan but for some reasons, this still surprises me 😭
Hahaha! 😂 I think it's because I really love JavaScript itself.
That's probably why React feels the most natural to me. You're working very close to the language, and I enjoy that a lot.
Angular would actually be my second choice because I like its more structured, class-based approach and the clear separation of responsibilities.
With Vue... I always have this feeling that too much is happening inside a single component. 😅
But that's purely a developer experience preference, not a statement about the framework's quality, performance, or what other people should use. That's exactly why I try to stay out of framework holy wars. 😄
These debates are fun🤭, but real business constraints usually kill the oversimplification. Once cost, delivery speed, reliability, and maintenance matter, the “best” technology becomes highly context-dependent.
Exactly! 😄 It's always nice to read a thoughtful take like this.
In real projects, things become much more complicated. Then the business comes along and says, "We're building this in Angular because of our existing ecosystem/team/client requirements," and... that's the end of the discussion. 😅
Even if Angular wouldn't be your personal first choice, it's not exactly worth starting a holy war over. At the end of the day, delivering value to the business is usually far more important than proving that your favorite framework is objectively the best.
With the kind of gigantic mess our tech stacks and frameworks have become, is it even possible to say anything with a tiny bit of assertiveness these days, let alone exuberating confidence!
If the core internet protocols and specs were simpler and designed properly, we wouldn't be needing frameworks and npm monstrosities at all to build apps and sites today. At some point, I think forking the protocols and building everything from scratch would be a more sensible decision for everyone than carrying the legacy baggage forever.
Exactly! Sometimes I get the feeling that the louder someone shouts, the more they're perceived as an expert. I've lost count of how many times that kind of overconfidence has led to terrible technical decisions, or at the very least, burned through an incredible amount of time and money.
Ha great article. There's a lot of performative AI out there and yet, I have a soft spot too and am always on the look out for tips, tricks and the next new thing even though tips, tricks and the next new thing may not necessarily be useful for my build journey. lol
Oh yes, me too! 😂 I absolutely love new technologies and experimenting with them.
But at the same time, I've learned that sometimes the best tech stack is the boring one, the one that's well-tested, stable, and everyone on the team understands. 😄
Yes, true!
The funniest part is that one-video confidence often sounds cleaner than real experience. Real experience has caveats, scars, and weird exceptions. The danger is not beginners learning from YouTube; it is mistaking a good explanation for operational understanding before the thing has broken in your own hands.
Thanks a lot for this comment! I completely agree, and reading it made me realize that this idea didn't really come across in my article. I focused more on the technical examples and the jokes. 😄
But that's exactly it. Ironically, the most confident people are often those who've just watched one YouTube video, or those who've only ever worked with a single technology and confidently declare it's the best one without ever trying the alternatives. 😂
Real experience tends to make you more humble because you've already seen enough edge cases, failures, and trade-offs to know that almost nothing is universally true.
That is a strong addition to the article, honestly. The joke works because everyone recognizes the pattern, but the deeper lesson is humility as a signal of real exposure. Once you have seen enough edge cases, you stop making universal claims quite so loudly.
Hahaha, I have a feeling you've just given me an idea for another article. 😂
That would be a fun follow-up. The joke has legs because every developer has felt that dangerous phase between “I watched one explanation” and “I have seen enough edge cases to be careful.”
This is so real. I feel like 'just add RAG' is the new 'just use microservices,' everyone says it like it's a checkbox and not an entire discipline. Also welcome back, hope Croatia was worth the FrontKon lie 😂
Hahaha, exactly! 😂 I think we've all been there at some point.
And yes, Croatia was definitely worth it! 😄
Not wrong; LLMs are just really complicated next word predictors.
I mean, NodeJS is just a while loop that imports a JS runtime.
In a slightly more abstract sense, your CPU is also just a while loop implemented in hardware.
I think the real problem here is people thinking "just a while loop" automatically makes it silly, when in reality, lots of cool things look like "just a while loop" from the surface, with most of the complexity going into making the inside of the loop work right.
"Any observed statistical regularity ..." and all that
This one is particularly baffling. Like, at what point do people lose the ability to just look at the big picture, notice that a bunch of tiny units are being spread out over a network and think "hey, we could just put all of this into one process"?
I really like that comparison with Node.js and even the CPU being "just a while loop." 😄
It actually reminds me that you can reduce almost anything to a very simple description if you abstract away enough details. In the same sense, you could also say that a human being is "just water and proteins." 😂
This had me laughing the whole way through. The CQRS one hit hard, I've definitely been that person nodding along to a video, only to stare at my own reducer logic a week later wondering what possessed me. And the AI agents being a while loop? Brutal but fair, until three weeks later and suddenly it's a whole ecosystem.
The unit tests line is painfully true. 100% coverage and zero confidence is such a specific kind of tech debt that nobody warns you about.
Microservices for six screens is the kind of overengineering I've seen way too often. Sometimes a monolith is just fine.
Honestly, I think we've all been that confident guru at some point. The trick is learning to laugh at it. Enjoy the next conference, sounds like a blast.
Haha, thank you so much! 😂 Your point about the monolith reminded me of something else. A healthy monolith is perfectly fine. Healthy microservices are perfectly fine too.
The problem starts when people build them without really understanding why... and somehow end up with a distributed monolith instead. 😅
That's probably one of my favorite examples of how adding more architecture doesn't automatically make a system better.
I honestly thought I was the only one in this boat. That zeal and drive that comes after watching a video or even reading a book that inspires you is crazy 😂.
I always wonder why it ends up that way when it comes to the implementation part..
Hahaha, definitely not! 😂 I think most of us have been there at least once.
Good to know. That's why a community is good. You get to hear of other people's story
Such a refreshing and realistic take on the "silver bullet" mindset in tech! The point about 100% test coverage with 0% confidence is pure gold. Huge congratulations on the iJS Munich lineup! That's a massive win. Looking forward to your article next week. Have a great post-vacation start! 🙌
Aww, thank you so much! ❤️ That really means a lot.
I used to follow Uncle Stef on YouTube, but stopped recently. For various reasons, but chief among them: AI, AI and more AI. That's everything he can talk about nowadays, like probably 90% of content creators in the coding YouTube video space right now.
It is boring and annoying. It's as if with the arrival of AI, normal tech developments have come to a halt and everything that matters now is agents, agents and everything agentic. Never mind that all these agents are based on LLMs, and thus only as good as their training data. Without new code being written by humans, how is the LLM ever gonna be able to generate code for new frameworks, or even new major versions of frameworks which aren't as well documented as for example Spring boot?
This "coding is dead" frenzy makes zero sense if you look at it that way. But it's still all any content creator on YouTube can talk about. And even if coding were dead, you still have plenty of situations where you need to understand the code the AI wrote, so specific framework skills matter now even more than ever before, precisely because we're switching more and more to reviewing AI code.
How are you gonna effectively review Spring Boot code if you know nothing about Spring Boot, to pick just one random example? It just makes absolutely no sense, and you get this stuff from people who have been in the industry for over 20, sometimes over 30 years.
Alright, I'm done venting now.
I completely agree. In fact, I wrote an article about this a while ago. The question I asked was: what if today's LLMs had somehow existed 15 years ago?
Would we have modern web development today... or would we still be happily generating index.php files? 😂
That's one of the reasons I don't believe coding is "dead." AI is incredibly good at accelerating what already exists, but someone still has to explore new ideas, build new frameworks, and push the ecosystem forward in the first place.
This is painfully accurate. The Dunning-Kruger curve has a "YouTube tutorial" bump on the front end that nobody warns you about.
The one I'd add: AI agents "understanding" code. Watch a 10-minute demo of an agent reading a codebase and it looks like genuine comprehension. In reality it's very good pattern matching over the specific examples it was trained on. The confidence is learned from the tutorial creator's confidence, not from ground truth.
The interesting thing about the tutorial-to-overconfidence pipeline is that it now has a second-order effect: developers who learned from YouTube are now writing blog posts and tutorials that are teaching the next cohort the same confidently-wrong patterns. The signal degrades through each relay.
What's the 8th thing? You cut off at 1-7 and I'm genuinely curious what made your list.
That's exactly why I like writing posts in this tone. They're meant to be a small counterbalance to all the "This is the only right way, just do it like this" content.
When I was a junior developer, I fell for that kind of "expert advice" more than once. It took me a while to realize that in software engineering, context and trade-offs usually matter far more than absolute rules.
So if my posts make even a few people stop and think, "Maybe the answer is a bit more nuanced than this," then they've done their job. 😊
The number of these that map directly onto explaining what an AI agent "decided" after watching it iterate twice is uncomfortably high. Confidently narrating a black box after a short demo isn't unique to junior devs, it's just a very human response to any system that looks smart.
Absolutely! In fact, I'd argue it's even worse when it comes from someone who's considered a senior. 😅
Day 1: Watched one YouTube video about Docker.
Day 2: Explained containers to everyone.
Day 3: Spent 4 hours figuring out why my container wouldn't start.
Day 4: Watched another YouTube video.
This post felt less like a joke and more like an attack on my entire learning strategy. Thanks for the laugh! Nice Read ❤️
Hahaha, I'm crying. 😂
That sounds way too familiar. My experience with Nx looked almost exactly like this. 😅
I guess some learning curves are universal. We all start with "Wow, this is so elegant!" and then reality politely replies, "Now let's spend the next four hours debugging." 😂❤️
I’d add one more: "Observability will tell us everything." Then the first production incident happens and you realize the logs are perfect, the traces are perfect... and the agent still confidently did the wrong thing. The system was healthy; the behavior wasn't.
Hahaha, oh yes, that one fits the list perfectly! 😂
Observability tells you what happened. Figuring out why it happened is often a completely different story.
Yuppp, exactly. That's one of the reasons I started building Failproof AI. Observability is essential, but we found it needed to be paired with runtime validation and guardrails not just explaining failures after the fact, but preventing them in the first place. Please check out our open source repo of Failproof AI(github.com/FailproofAI/failproofai). Don't forget to hit the star if you find it helpful ;)
Hi @sylwia-lask , you represent the pure embodiment of the developer I've always imagined and envisioned. A developer not only needs to be intelligent, but to survive in this industry you also need to be ironic, sometimes sarcastic, and definitely equipped with a healthy skepticism.
Reading your posts, I see myself again with many less years. In my day, there was no YouTube, just publications, manuals, and a few (very few) flesh-and-blood "gurus" who explained things face-to-face, and therefore had to be careful what they said.
Then there were specialized forums—C, C++, Java, PHP, etc.—where each section extolled the qualities of its own language to the detriment of the others. Ninja battles. Today, I also spend time watching YouTube, but honestly, with great detachment, because the superficiality with which the topics are treated is on a par with that of telemarketing salesmen.
I just have to acknowledge their remarkable ability to ride the wave and capture the attention of the masses. But a professional knows full well that this is just the tip of the iceberg, with a whole hidden world beneath.
I've always told my students, "There are no easy or difficult things." There are things you know and can do well (easy) and things you don't know (impossible).
So thank you for these posts, which always bring a smile to our faces.
Good luck!
Thank you so much for this comment! 😊 I especially loved the part about gurus you had to meet face to face.
It reminded me of my first job, where I had one tech lead whom I treated almost like an unquestionable authority. Then another tech lead joined and started openly criticizing—sometimes even mocking—many of his decisions. And honestly, he had a point more often than not. 😂
That was probably the moment when my blind faith in gurus ended for good. Since then, I’ve tried to listen carefully, but also verify, compare perspectives, and remember that even very experienced people can be confidently wrong.
Number 4 hit way too close to home! 😂 I started my AI agent journey thinking, 'Oh, it’s just a prompt in a loop,' and three days later I’m knee-deep in function calling, error handling, and trying to figure out why my agent decided to hallucinate a new API endpoint. The 'YouTube tutorial vs. Reality' gap is real!
Hahaha, that's exactly how it goes! 😂
I've definitely watched at least one when starting out. That's why I believe it's important to do your own research and not just fully believe everything. Form your own opinions after, everyone is different after all. Also applies in life as well. Trust but always verify.
Oh yes! 😂 And I have to admit that failing a few times after following the "one true, obvious solution" is a surprisingly effective teacher.
Nothing builds healthy skepticism quite like discovering the hard way that what looked perfect in theory doesn't quite survive contact with a real project. 😅
"Trust, but verify" is probably one of the best principles in software engineering.
the while loop item makes me laugh most because the three weeks later version is exactly what happens. the loop stays simple — the complexity moves into the tool contracts.
once you have multiple MCPs with overlapping capabilities, the hard part isn't whether the agent loops but which tool it picks, in what order, and what happens when the first tool fails partway through and the second tool's context is now stale. that's not a loop problem, that's an orchestration problem, and the answers are mostly still empirical. is that the kind of complexity you ran into with the agent work you mentioned?
Exactly! That's been my experience as well.
I think that's also why we're suddenly seeing so many conferences, talks, and articles about AI agents. If it were really "just a while loop," there wouldn't be so much to discuss. 😂
The loop itself is usually the easy part. The real complexity comes from orchestration, tool selection, context management, retries, failures, memory, and all the interactions between those pieces.
It still feels a bit like the Wild West. We're discovering good patterns as we go, but I don't think the industry has settled on many universal best practices yet.
Excellent article
Number 5 could just as easily be "WebSockets" ha ha
Oh yes!!! 😂 If I got one cent every time I heard that, I'd already have... at least a dollar by now. 😄
HAH you picked the nial
hahaha thanks!
nice
Thanks!
I do use YouTube to listen to great music, watch movies but not for learning anything
haha same here
Interesting!
Nice post!
Welcome back from Croatia 🙂
Hahaha thanks Ekong, I missed DEV!!!
Haha, this is so relatable. One YouTube tutorial and suddenly we feel like experts in a completely new technology. The confidence curve is always faster than the learning curve. 😂
Oh yes! 😂 And then came the implementation... and suddenly it wasn't quite as straightforward as the tutorial made it look. 😅
I think we've all been that person at least once, Sylwia. The more experience you get, the more you realize almost every "always" or "never" in software should probably be "it depends." Great read 🙌
Oh yes, absolutely! 😂 I think every one of us has been that "guru" at least once after watching a great video.
😄😄😄
Thanks
😊
I remember checking youtube videos and thinking everything is so easy, and later stuck at basic steps
Hahaha, exactly! 😂 I still remember watching my first Angular tutorials and thinking, "Wow, this is so easy!"
"Let's just add RAG" hit way too close to home 😂
I actually just went through this while writing a beginner tutorial on it the code to wire up a basic retriever + LLM chain is maybe 20 lines and takes 10 minutes. Nobody warns you that the actual work is chunk sizing, overlap tuning, and realizing your "closest match" by embedding similarity isn't always the semantically relevant one. That's where the real hours go, and it's exactly the part no YouTube video covers because it's not demo-friendly.
Same energy as #6 honestly - RAG demos look like magic in a 5-minute video for the same reason AI-generated tests hit 100% coverage: they're both optimizing for the metric/moment that's easy to show, not the part that actually holds up in production.
Great list — painfully accurate on #4 too. Every "simple while loop" agent I've built has mysteriously grown an orchestration layer by week three.
The Dunning-Kruger effect is incredibly prevalent in web development right now, especially with the explosion of AI tutorials. I have seen so many developers confidently implement complex state management or database schemas based on a single ten-minute video, only to create massive performance bottlenecks later. The real challenge is that video content often omits the edge cases and failure modes that only reveal themselves in production environments. How do you usually handle mentoring developers who fall into this trap without discouraging their enthusiasm for learning new tools?
That's very relatable! One YouTube video can introduce the basics, but real expertise comes from building projects, solving problems, and learning from mistakes. Continuous practice is what makes a great developer.
CQRS is easy... Yes, when folks miss it for CQS
fox? you are hard to catch. dglassesguy gmail. dollhouses galore. please. all built on your work. 8 sites ready to go. blow their minds. call me.
Great analysis! The comparison table is really helpful. I've been exploring similar tools lately.
Great analysis! I've been exploring AI coding tools too. What's your take on open-source alternatives?
Great article!