It is 2026, AI is everywhere in coding, particularly at my job. AI says what to build, AI says how to build it, AI builds it, AI reviews it, AI fixes errors, sadly AI can't go fight the client just yet. And as managers & bosses realized that a simple $20 suscription can pump out more code than ever, naturally schedules shifted.

Now using AI is the expectation. If before a simple CRUD took like, idk, 4 hours because of all the "enterprise" boilerplate, now AI does it in 10 min tops. And AI generated code is now "good enough", that people (me included) can just not even code anymore, but just prompt.

But that got me thinking: did AI make me no longer enjoy coding?

Coding as a hobby

There is a lot of people in the industry, and I would say most of them are in it for the money. Which is fine, gotta pay them bills. But me? I am in it because I enjoy it.
I enjoy the process of solving problems, the creativity of coming up with solutions, the process of weighting & considering tradeoffs & balances. I enjoy the little things of programming, from the syntax of what I write, to the big & slow deployment pipelines & servers.
And when AI automated so much of my coding, it got me thinking: did I just lost my hobby?

Dumb analogy

It felt as if my hobby was woodworking, like, I have my tools and my wood and would enjoy spending the weekends working on a new closet. But then, a new IKEA opened near me, so, my hobby has now been replaced by: buying a closet from IKEA.

In a similar way, if I enjoy coding, and I let AI do all of the things, I am no longer practicing my hobby. Now its as if the AI is the one with the hobby, and I'm no more than a simple bystander. The realization of this hit me, for I was now using AI for the projects I do on my spare time, as a "hobby". I saw the AI write mountains of code, creating a full project in record time, and make it work more of less ok. But something was missing.

Missing stuff

I was missing the enjoyment of seeing something I'd done, that only comes after spending tens of hours on a project.
I was missing the joy of seeing a small interaction perfected, through trial and error.
I was missing the pain of having a really weird error, solving it after a 4h session of trial and error, and in the process having learned a ton of things.
I was missing the process of having no idea of how to do a thing, but doing it anyways, my way, and then look up what the "proper" way is.
I was missing the knowledge to even evaluate if the "proper" way fitted the project.
I was missing the connection with the project. It wasn't really my project. It was the AI's. All I did was give vague directions, tell it to fix that, to use these colors, to have a specific flow. But it was not related to me in any way.

I vibecoded a personal finance app. I just had an idea of what I wanted, and told the AI to build it. But I had no idea of anything that was going on. Maybe not even the idea was mine, I would ask the AI what would be some good improvements over and over. And those must have came from some training data, which come from some real projects pre-AI. Someone I don't even know the existence of decided to do something a certain way, and the AI just autocompleted with that. It was not my idea, it was not my db, those were not my operations, that was not my UI, those were not even my bugs.

Realization

I realized a lot of things about AI lately. Mainly, I realized that my hobby was like woodworking, and that I get joy from the process, not only from the finished product. But I also realized the ways that AI is atrophying my brain.

I was no longer thinking for myself. Since the AI was "good enough", I let it take the decisions, write the architecture, build the feature, fix the bug. I was no longer learning. At work we are moving from TS to C# for REST APIs, and introducing many enterprise patterns, like, CQRS, Event Sourcing, Microservices, etc. And the AI was doing it all. I had no idea how those things worked, how they fit together, what were their merits and drawbacks, when to use them, when not to use them, nothing. But, since the AI made a plan, gave a list of pros I didn't even bother to fully understand, I went along.

Taking pride on organic thoughts

And really, if tomorrow the AI tells me to "do stuff like this", how could I counter? How could I say that it was right or wrong? With what knowledge? Do I ask a different model to counter it?

And what will happen when the model makes a wrong decision? When it doesn't see the whole picture? Do we just pray that AI becomes good enough, and fake it until then? Do we pray the AI doesn't go rogue? Do we pray providers don't run out of VC money, the bubble pops, and prices rises 20x? What will happen then?

Someone has to have the knowledge, to judge whether the AI is producing good or bad results. The AI can't be its own judge. And that knowledge should be held by a human. Someone whose brain isn't limited to 200k words, that needs a summary after that, and inevitably loses nuance. Someone who can say if it is right or wrong, based on years of real experience, and doesn't need handholding or explicit instructions. Someone with strong opinions, and not just a yes man.

Because if all you are is a vessel for the AI to speak, if all your thoughts are limited to what the AI tells you, if you need the AI to tell you the pros and cons, if you have no ideas of your own, what is your worth? I would say, you are worth as much as how many dollars your Claude Code suscription costs.

---

So, at work? It's joever. Management now expects you to use AI and 10x that productivity, and they expect you to leverage the AI to improve things way outside your skill set. Engineering is dead.
But at home, for my own enjoyment, I will be building my own closets, beds, tables, chairs. With my own hands, my own tools, at my own pace. I won't be buying them from IKEA anymore.

Just some thoughts I'm having. Everybody has their own definition of these things, for some reason.

Vibe coded

AI completely drives all. You just say "build the thing", "fix the thing", and whatever garbage comes out of the AI, that's the project
"Claude, build me a AI chat app, with support for OpenAI compatible providers. Do it in react with Tanstack Start"
don't care at all about.

AI first

AI drives the main execution, but the human now has a more involved plan. The human knows the project, has ideas about how things should work, how things should look, sees ahead, plans broadly. The human commands the AI to execute the plan, supervises, changes course. More involved, but the human is more likely to know what the fuck is happening, if only slightly.
"Claude, write the input box for the messages to be sent to the backend, for the AI chat app. It should: - add newline on enter; - reject empty content; - allow pasting images; ..."
I'd use this for ideas i want to quickly test/validate. Unfortunately, nothing is more permanent than a temporary solution...

Human first

The human is the main driver of the project. Human has idea, plan, details, and executes it themselves for the most part. AI helps in writing boilerplate, writing code that the human already knows in and out, but that they're past. The AI is like a junior dev, just a helper for the senior dev, who is preoccupied with carrying out the hard part. I would consider the human knows what is going on, sans tiny details. But the human can just read those and immediately understand them.
I think this would include basic TAB complete to small, discrete tasks by an agent. I'd use this for things I really care about, and where I have enough expertise (I've done it before, know what to do, etc).
"Claude, wire the state of react-hook-form form to the input els, handle events, handle focus loss."

Organic

All code is 100% human written. The human knows all of what is happening in the system. It is very, very slow, but more likely to produce a good result.
For things I really care about, things I want to learn, and that I am not an expert on/haven't deal with before. Like, idk, writing a compiler and Virtual Machine and Garbage Collector and Allocator and stuff.

Yesterday Anthropic Mythos was announced, and everyone is panicking. Apparently it's capable of hacking into literally any system? And find zero days hidden for years. It looks like a really powerful model has arrived... yet, this was said about almost all models. Furthermore, people have been claming that AI is capable of replacing humans for some time now, we have Opus models and the latest GPTs. Yet I must be doing something wrong, because sure they are able to do stuff, but only in tiny projects. And even then, they take their sweet time and sometimes miss the mark.

Skill Issues?

I certainly lag behing as far as AI goes. When ChatGPT became the new hot thing I was still hand coding. When Agents happenned, I was (sometimes) using chats & tab complete. When MCP became a thing, I was just using agents. And nowadays there's skills and souls and subagents and ralph loops and so much stuff, it's hard to even know what works.

AFAIK the hottest new thing right now is spec driven development. I've been told that OpenSpec is a turning point in AI. Yet when I try it... it just spends some hundred thousand of tokens to create a plan, produce artifacts, create md files and all those wonderful things... then other hundred thousand of tokens to implement... then other thousand of tokens to store... and I'm just wondering: can't we just implement the thing?

Other things I tried are skills, tools that keep context (engram), rules. None seem to work for me.

It would seem that I just don't get it. A lot of people talk about vibecoding on the internet, yet when I sit and tell the AI to "do X stuff", it does it sure, but more often than not the output is... questionable. And when it builds upon previous AI work, things get messier and messier.

I also think a lot about how, if vibe coding is so good and coding has been solved, we would've seen perfect (or better) clones of popular services spawn in no time. Like, you tell your AI (or ralph or whatever) to build a Slack clone, deploy it, then go earn money. Jira connection? GitHub connection? Just tell the AI to build it right? Why hasn't that happened yet?

It would appear that even when the problem is so well defined and documented, AI is not capable of autonomously building big stuff? Or is it just me not looking around? I see GitHub became one of the worst platforms, I see Win11 becoming a meme due to the amount of AI introduced, I see a lot more CVEs are found, new vulnerabilities every week.

If someone knows the secret sauce to tell the AI "go build me a distributed sync music player", tell me, for I am not able to have it working.

So over the last weeks all major AI platforms are tightening their rates. OpenAI will no longer offer discounts, Anthropic is enforcing tighter usage limits & removing Codex from cheap plans, GH copilot just announced a move to token based. Even regular API prices are increasing, the latest GPT 5.5 is now even more expensive than Opus! It's so joever!

And that got me thinking: Are we about to witness the end of AI vibe coding? Like, without all those services subsidized, we will actually see the **real** cost of AI. And it was just these last months that I've noticed that the use of AI increased. Like, even a tiny prompt might trigger 5 subagents each scanning the whole codebase! (but that is def a skill issue 💀💀). All those tokens will now translate to raw dollars... Like, right now a single prompt with Sonnet 4.6 in Cursor costs me like $0.50.

With the death of vibe coding as we know it, there'll definitely be a decrease of clueless people just throwing messages around. People might actually need real, hard skills to build complex systems! (or be like, so rich).

On one hand, I'm happy that real skills will be needed again. But on the other, I'm fearful that I've let AI do so much of the work that, some of my skills would no be athrophied and no longer be enough. Like, at my work they will definitely not be paying those API prices. So here's for more real engineering & thinkering!