I know this might sound a bit hippie coming from me, but the last couple of years have been crazy.
The first time I tried ChatGPT
I was among the first people to try ChatGPT when it came out, mostly because I spend an unhealthy amount of time on Twitter (sorry Elon, I will never call it X). I was already into AI stuff thanks to the GitHub Copilot preview, so I was fascinated by the idea.
I was a freshman back then, so I started testing it with Calculus and CS questions. It did terribly.
The next day I had a lab practice for my Programming Fundamentals course. I told my professor about it, she was as excited as I was, and we gave it the lab instructions. It did a decent job in less than 30 seconds. We were shocked: a Python lab meant to take about 15 minutes, done in under 30 seconds.
Of course, it wasn't perfect and it made some mistakes, but as soon as we told it what was wrong, it fixed them. Deep inside, I knew I was cooked. I had spent years studying to be decent at programming, and this new tool just did it in seconds? tf?
My denial phase
So I entered my denial phase, saying stuff like "yeah, I mean, it's cool, but it will never be as good as me at coding, I have years of experience."
Then I went to the library to do some Calculus homework. I was solving integrals left and right and thought I was pretty good, until I hit the integral of sec³(x). I spent a long time on it and finally got the answer. Then a classmate asked, "why didn't you just ask ChatGPT?"
So we did, and the answer was a complete mess, a bunch of steps that made no sense. I looked at him and said, "see? This thing can't think. That's why we're studying engineering."
From that moment on I was pretty confident: it was cool, but I'm studying engineering. It can't be smarter than me, right?
Always early, never convinced
Months passed, more people started using ChatGPT, and it became the new big thing. Its problem was that it hallucinated a lot: if it didn't know the answer, it would make up straight-up bullshit.
Then Perplexity came out with a simple selling point: "it doesn't invent answers, it gets them from the internet." A huge milestone. I knew it was "just" ChatGPT with internet access, but I started using it and the results were considerably better, even if you could still feel ChatGPT underneath.
Then Cursor came out in 2023 and people on Twitter went crazy. For me it wasn't a big deal, because I had been using GitHub Copilot since its preview in 2021. Honestly, I thought it was a dumb idea: why a fork? Why not just a VS Code extension like Copilot? I kept reading about the new features Cursor shipped every week and still wasn't interested, because I "already had Copilot."
Then Cursor gave students a free year of Pro, so I finally made the switch. I was fascinated. Cursor Tab was way faster than Copilot, you could plan, switch modes and pick models. It was amazing, and I felt dumb for not switching sooner.
See the pattern? I was almost always an early adopter, but I was also convinced I knew better than the people on Twitter, so I never jumped straight to the new thing of the day.
Then it got better than me
Time passed, new models were released, and Cursor kept getting better, until Cursor Tab was doing most of my coding. The agent still wasn't good enough, though: I would give it a long-ass prompt and it would do whatever it wanted except follow my instructions.
You probably know the rest of the story: then came Claude, then Claude Code, and so on. The thing is, now I know models are better than me at coding, the very thing I spent years studying. And even when they get something wrong, they fix it way faster than I would.
I still review the code they generate, but it has reached a point where I barely have any comments, so I'm probably wasting some time there.
Software isn't dead
I know this story is very software-engineering-centric, but we are the most visible field and the perfect example: we are the profession that was so obsessed with automation and optimization that we automated almost every part of our own jobs, haha.
And don't get me wrong, I'm not saying software is dead. It has evolved, and there's almost no barrier to entry anymore. Anyone can build and deploy a website in seconds, and that's the beauty of it. The hard part was never the coding. It was solving actual problems, then maintaining and scaling those solutions.
There's something else people forget, too: software is not just websites.
Steps and loops are machine work
As a society, we have built a system based mostly on steps and loops. Those are the perfect use cases for AI. If your job is following the same steps every day, you must look for a way to hand that workflow to AI.
You are a human. You don't need to spend your best years going to an office every day to do the same thing for the rest of your life. That's machines' work. That's why we built them.
What college should teach now
We should also thank AI for exposing a problem: we send people to college for four years (or more) to learn skills they mostly won't use at their jobs. Now anyone can learn anything, and the answer is just one prompt away.
It's now evident that programs should focus on the human part of each major: how we relate to and communicate with people, how we sell, how we identify a problem, how we model it, how we maintain and scale a solution, and a bunch more.
For office jobs, a diploma doesn't really certify anything anymore. Of course, some careers are naturally different and this doesn't apply to them.
Technical knowledge still matters
Does this mean technical knowledge is obsolete? Absolutely not! It's more important than ever, and this pattern was around long before AI: the more you know about software engineering, the less you actually code.
Coding is the plumbing part of the job. In the end, we always need a human with solid knowledge and experience making the decisions. What might change is how we get that knowledge and experience.
And don't forget that humans are the ones building AI.
You are the boss now
The way I see it, this isn't very different from what we have today. Have you ever had a boss who was dumber than you? I have some news for you: you are now that dumb boss.
So what kind of boss do you want to be? The one who's only there because he's the owner's son, or the one you learn from, admire and respect?
To be the one making the decisions, you have to have risked capital, fixed a problem, built the solution and sold it (or just won the nepotism lottery, haha). At some point, someone had to take the risk, act and, most importantly, convince people to follow their ideas. Agents fix the plumbing now, so be the boss and convince people to follow yours.
Does this mean everyone should be an entrepreneur? Nope, but it's easier now than ever. Use that leverage to fix problems people have and make their lives better, and people will reward you for it. If you'd rather help others build and maintain their solutions, that's cool too. Just don't stay in the plumbing. That's machines' work.
Demanding a human for machine work
There are things AI can't do, and some people will demand to deal with a person instead of an agent. That's OK, friction is expected. But what are we going to do, go back to doing every interaction in person just to make sure we're talking to a human?
The problem isn't agents. The problem is people refusing to let agents do the work and getting in the way of automation. Humans should step in before or after, to make the decisions.
And for the skeptics: what happens if I put you on a call with an agent that sounds and feels just like a human? You get the same output, whether that's a sale or a meeting, except an agent did the work and you thought it was a human. The output is what matters. Insisting on a human there is demanding a human for machine work.
AI isn't free
There's something people almost always forget to mention: AI is not free, or even cheap. It costs money, and a lot of people and organizations can't afford to implement it. So it's not an equal competition.
But that's just life. It has always been like that, and there's nothing wrong with it, as long as competition keeps pushing costs down.
So what's left for humans?
If most office jobs are machine work now, what's left for us? What are "human jobs"?
Identifying problems, solving them and making the world a little better every day. To do that, you need knowledge and experience, which essentially means learning. That's the human job: constantly learning about everything around us.
I know people will get tired of it. Some will learn way faster than others, and some will spend years learning the "wrong thing," but I guess that's just life: the more you fuck around, the more you find out.
Go be human
AI is here to automate the boring stuff, make us more productive and generate value. Now is the best time for you, as a human, to pick up a new sport, hobby or activity that makes you happy.
You have access to knowledge and digital execution right from your pocket. You can have agents working for you while you're on a road trip to the countryside, while you're hiking, while you do anything that makes you feel alive.
I know it's going to take a lot of time, and a lot of people will suffer in the process, but believe me, the future is exciting.
The lesson
Sure, I'm oversimplifying jobs, the problems people face every day and how hard it will be for the world to adapt. But as you've seen, I was wrong about what AI could do for years, and that taught me something life and history keep showing us: don't resist change just because you're afraid. Try, evaluate, fail (I did, everyone does), and keep trying.
The world is changing, and it always has been. It's normal, and human, to be afraid of change, of new challenges and of things you've never dealt with before.
But the solution is not blaming AI for "destroying jobs," or trying to ban or regulate it. It's embracing the change and learning to use these tools to make our lives better.
As the quote often attributed to Marie Curie goes: "Nothing in life is to be feared, it is only to be understood."