Should you learn to code in 2026? A letter from someone who entered a trade ten years ago that has since become a different trade

Day one

If you are reading this, you probably have two contradictory things in your head at once: that AI is taking programmers’ jobs, and that this is still one of the better trades you can pick. Both are somewhat true. I will not tell you which one more. I will tell you how it went for me — because I entered this trade ten years ago, and the trade has since become a different trade. Same name, different work.

My first day in an office job stressed me out. It only let go when I sat down at my first desk. I had a mentor — deeply human, empathetic, and at the same time a beast at programming. That combination is rarer than it should be, and today I know how much I owe him.

The first surprise came fast: I could not set up a working environment on a new machine as easily as I had imagined. Days passed before I wrote my first line of code. Not hours — days. Today that would be an hour or two with an agent. Back then it was torture.

I am not telling you this because it is a funny anecdote. I am telling you because the fear you feel today — “I can’t do this, everyone else knows more” — is exactly what I felt at that desk.

Three months and one sentence

I was afraid I was not ready. And I had reasons: I had been learning to code for just three months, from zero. I did not have much of a clue.

What I had were three things. A very well-built plan — based on research into the job market, not on what happened to be fashionable. An open mind. And a huge amount of motivation, which let me not think about the obstacles and simply act.

Those three things produced three months of diligent study and one working project, which I showed at the interview. Remember those three things, because I will come back to them at the end. The market has transformed since then, and I believe they are as valuable as ever.

The fear did not come true. Not because I knew enough — I did not. I lost a huge amount of time wrestling with tools I did not understand; docker compose was black magic to me then, and today it is something I do not even notice and cannot imagine working without. The fear did not come true because this trade is peculiar: what you know does not win. What wins is the ability to keep absorbing new topics, logical thinking, and an open head.

That is the one sentence I would like you to keep from this letter, even if the rest fades.

How AI changed my work

Until recently everything looked the way you imagine. I wrote all the code by hand. I tested by hand. I hunted for the causes of bugs by hand. It was absorbing work and I kept getting better at it. Until everything changed.

The first time I launched Claude Code in a terminal — still in beta then — a few days of playing were enough to open my eyes: the times I work in were about to change enormously. At first I was sure I had found the holy grail. A few weeks later I was writing my first client project in it.

And here begins the part you will not read in most posts about AI.

The project went into testing with real users — the client’s employees, because it was a tool for them. The client agreed to that knowingly, as an experiment. And the experiment showed how early in this revolution we were.

The application’s main loop was badly tangled. The model, given instructions, tangled it further. It went like this: the client relays from the users what does not work. I relay that to the model, together with how to fix it. The model fixes one thing and breaks another. And again from the top. The client was getting annoyed, so was I. The ping-pong went on for days, maybe weeks.

With hindsight I can see what went wrong: the design process was wrong from the start, because there was no clear plan for the project, it just grew brick by brick. The model hallucinated a lot back then. My instructions were unclear. But I could not see that at the time. All I saw was that it would not fix it.

So I stopped asking and started reading. I dug into the code, analysed it, wrote the fixes myself, the way I had for all the preceding years. It was a moment of doubt — but also the moment I understood two things at once. First: an agent cannot be left unsupervised, because it will make a mess. Second: to supervise it, you need experience and understanding. Those ten years were suddenly not ballast. They were the only thing keeping the project alive.

Then better models came out, and I decided to give the agent a second chance. And at last I started getting what I needed. By then I was certain it was only a matter of time before everything changed. While others doubted, I used it.

Hands and head swapped places

Today my work is mostly conceptual. It used to be that most of the day went into writing code, and thinking about what we were building and why happened in the gaps between one line and the next. Now it is the other way round: most of the day is decisions, plan, criteria, reading what the agents did. Writing code by hand is a margin.

What I am responsible for has not changed. I still decide what “correct” means, I still answer for what reaches the client. What changed is what I do with my hands. And if you are wondering what to learn, this is where the answer is: learn the part that stayed in my head, not the part that went into the agent’s hands.

Is it worth learning to code when there is AI?

This is going to be uncomfortable now, because I do not want to lie to you.

Getting into the trade is harder today than ten years ago. It is not just my feeling: according to the No Fluff Jobs report “Rynek pracy w IT 2025/2026”, juniors get 5.3% of the offers, one in twenty, while almost 60% of openings are for seniors and around 35% for mids. What is needed is someone who understands what is going on — less and less in the code, more and more in its architecture. And for that you need either experience, or an expert’s skill at driving models and recognising which of their proposals are actually good. Either way: experience. And experience cannot be generated.

Something follows from those 5% that no course will tell you: do not aim for junior. The junior role is dying, because the work a junior used to be given to learn on is now done by an agent. A simple task under a senior colleague’s supervision? That is exactly the job description of an agent. Fighting for one offer in twenty against hundreds of other people fresh from a course is lost before it starts. One offer in three is for a mid. Aim straight there: for someone who takes a task and brings back a working result, with agents as their juniors. Ten years ago that would have been cheeky. Today it is the only way in that is open, because mid is not a number of years on a CV but the ability to take responsibility for an outcome. And that is exactly what you can learn on your own projects, without a job.

The second uncomfortable thing: pushing code to GitHub and showing it off on your CV no longer makes sense. Code like that can be generated in minutes today, and the person reading your CV knows it. Nobody has time to read it, and there is no real point, since it is known to be written correctly. What counts is whether the project works. A portfolio today is a working thing that can be clicked and has a user, not a repository.

What counts is a way of thinking and an understanding of the new process of building software. And that, contrary to appearances, is good news — because it can be learned from zero, without ten years behind you. It takes practice.

What to learn in 2026, and what not to

If I were starting today, I would try it like this:

• Do not be afraid of topics you know nothing about. I have built projects in fields I had no idea about — and lately it happens that I do so without a single glance at the code. They are in production today — because an open head lets you absorb knowledge fast, and project experience lets you make the right decisions. Of this whole list, this is the most important.
• Learn to say what you want in a way that lets you check whether you got it. “Build login” is a wish. “After three wrong passwords the account locks for fifteen minutes” is a task, because the question “is it done” can be answered yes or no. My first project with an agent fell apart on exactly this: I gave wishes, I got guesses. Today I write every task to one template. It was written for the people who commissioned work from me, and it turned out to serve agents just as well.
• Build your own things with an agent — as many as possible. Not tutorials. Something that has a user, even one, and works exactly as planned. Direct the agent’s work, ask and probe why this way and not another, what the alternatives are, why this one is best.

And what I would not do: I would not buy courses that promise a job in three months. I know I made it in three months, but not thanks to a course — I did not spend a single złoty on learning. I would not compare myself to people on LinkedIn, because all you can see there is what someone chose to show. And I would not learn “for the interview”, because interviewing is changing too, and a year from now it will test something else.

One sentence to end on

I promised to come back to the three things from the beginning: plan, open mind, motivation. Ten years ago they got me into the trade with three months of knowledge and one working project. Today knowledge from a course alone will not get you in — but those three things still will. All that changed is what to point them at. The plan: at what the market is looking for now, which is an understanding of process and architecture, not another framework. The open mind: at the fields you will build in, because code has stopped being the hard part. The motivation: at your own projects with a real user, because that is the one experience that cannot be generated.

To finish, back to my mentor. I got a mass of knowledge from him that carried me through the first two years. But something else mattered more: a sense of belonging. I was accepted by people in the trade. I got empathy and good stories from someone who had already been there. You will not get that from any agent — and that is why I am writing this letter.

I am not promising you a job. I do not know what the market will look like a year from now, because it is even more volatile today than ten years ago. But one thing I will tell you: do not fear the unknown, and keep an open mind. Everything else can be adjusted along the way.

What next

This letter opens a series. I wrote in it that getting in is harder and that you have to aim higher, so now I owe you something: showing how. The next letters will be about specifics. How to build with an agent a first thing that has a user, and not end up with the tangle I ended up with on my first client project. How to break a problem into tasks an agent will do well. How to get a mid’s experience when nobody will give you a junior’s job. I do not have it all mapped out yet, and I will not pretend I do.

So if you have a question this letter does not answer — write to me. I will reply, and the questions that come in will set the next letters.