I Love My Job Again

I am drunk again most recently.
This time from drinking too much AI.

My interaction with AI went from using ChatGPT to GitHub Copilot agent mode. Just what I am experiencing recently made me love my job again.

AI Operating System

It was last year during Devoxx when I chatted with a friend. I didn’t want AI plugins or additions — I wanted an operating system interface based on AI.

I didn’t want AI over it; I wanted AI under it.

Apparently, I was talking about agents but didn’t know it then. The basic idea was to say confirm that email on my PC, instead of opening Gmail, finding, opening, clicking manually.

I wanted to deal with what, not how.

I even worked on a tiny prototype to interact with my computer that way. I could only build what I did not want to have. I stopped that adventure.

My head was bigger than my hands.

However, I kept using ChatGPT as a daily brainstorming partner.

From ChatGPT to GitHub Copilot Agent

For some years now I have a pet language-learning app project. I started to pick up interest again in working on that recently. I used to suck at building UI — frustrated by not being able to create what I imagined. ChatGPT helped me build some fancy stuff I actually was happy with. Last week, I was introduced to GitHub Copilot agent mode — and how it could do things for me.

So I fell into a rabbit hole.

It wasn’t long before I started writing requirements instead of code snippets. I design more than I code now. I review more, and write code only when I feel like it. I’m not lost in details anymore — I’m building what I want.

I can go from an idea to implementation, even to production — or to garbage — in a single day. I develop and try out features at unbelievable speed.

I can build solutions based on concepts instead of tools. I can design in interfaces instead of implementations.

My mind shapes my hands now.

I have only one hammer,
A hammer which knows almost all other hammers.

Intuitive Design

I’d like to share a story here.

I used to watch TV series on Netflix, especially while having lunch or dinner. Once, it was with a friend — and it was Friends. I used to skip the intro by using the touchpad (imagine the drama with oily fingers). I told her once, “If I were developing this app, I’d make the ‘S’ key skip the intro.” Then, naively, I pressed S — and it worked. That is intuitive design — where the minds meet.

Why I shared that story: I don’t know how to implement that feature, but I know that’s the design that feels right.

With coding agents,
I can end up with what feels right.
Instead of what I can do.

My vision is no longer handicapped by my hands.

High-Order Learning

I’m learning Dutch, while building that pet application using GitHub Copilot. At times, I’m not even sure what I’m learning.

Is it the language?
Is it how to work with AI?
Is it what my job really is about?
Is it tools, or technologies?
Am I learning Dutch?
Am I learning what a language is?

It’s like a playlist on loop. I go through them by what feels right. Design becomes messy when attention goes to content; content becomes boring when attention goes to design.

So it is a dance. I am a team.

Vim

I don’t know why, but my feeling tells me this is the right place to mention Vim.

When I found out about Vim, I loved it immediately (love on first sight).
It was challenging. Something so simple suddenly became complex — and beautiful. Vim is complex, but it is simple. That’s how genius hides.

Who could have thought that boring editing work could become a creative act?

It wasn’t about learning Vim; it was about getting to know the creative mind behind it. Vim is for minds that are faster than hands. Nothing can match the speed of creativity — we can only try. Creativity is not about how; it is about what.

What includes how. How becomes what.

As Vim changed how I sculpted my text, agents are doing something similar for my ideas. I just need to speak in the right keystrokes.

I know what :wq is — and agents know how :wq works.

Frameworks and Libraries

I’ve heard concerns that AI could generate “crappy code.” I’m not worried. That issue, in a different scale, was always there. We (ab)use libraries, frameworks, Stack Overflow — often without understanding abstractions.

I might not know how to implement a solution with Signals, but I can recognize the problems that Signals aim to solve. Recognizing problems is more important than implementing solutions.

Problems don’t change — solutions do.
Solutions don’t change — implementations do.

The JVM freed developers from memory.
AI will free people from their own memory —
so objects can be fluently created and garbage-collected.

The Third Space

I remember the first time I discovered HTML and how you could make web pages. I had limited internet then. I ran to my friend’s house with a USB stick to download a program — I assume it was an editor of some kind.

Finding out the capabilities of code agents feels the same. I’ve fallen into that same spirit pit.

I remember the bloom of Web 2.0 and social media. I remember when smartphones and mobile apps became the norm. Each opened up huge spaces for creative ideas. I was just sad that I couldn’t join them.

Now, with generative AI, I feel a third space has opened. It is the new playground — and I feel welcomed.

It is also confusing in a way:
Am I using a smartphone or am I developing an app?

AI Is the New CLI

Once, I got to know that there was something different from blue. I was as happy as a kid tasting chocolate for the first time — It was orange and purple. I thought computers were supposed to be blue forever.

There were other colors.

Blue had hidden the most beautiful world from me — a world I could talk to in my own language, instead of pointing at pictures like a child.

I wasn’t just learning the CLI —
I was touching the genius, the creativity behind it.
I was falling in love.
I was underage to be drunk then.

I Need a Junior

Years ago, I was trying to build some of my ideas, but I was limited by my hands. I remember telling a friend that I needed to hire a junior for UI work. Maybe even give my projects to interns or students — I needed execution power. But I didn’t have the money or the comfort to do that.

So I tried by myself — a to-do application (a complex one, I assure you). I started building: DigitalOcean, Terraform, containers, composers, React Native, database management, project workflows, and so on. It became learning-obesity.

It was messy. I was not a team.

After a while, I got lost. It was too much. I didn’t have that much brainpower. That kind of higher-order learning was above my level. My love and energy were limited.

Now, I have smarter, faster, and more knowledgeable partners.

Thank you, everyone for contributing in pieces to this wholeness.

I also remember recently wishing I could work with artists — so they could help bring some of my vision and ideas into reality. For example, ideas for short movies, illustrations etc.

It is interesting how creativity works, it is not even in one single field.

It takes 10,000 hours to become an expert — how much time is spent training AI models?

I am a Junior

I can imagine some of us feeling threatened by AI. One might feel they’re being told what to do — through code completion. Another might feel insecure that AI can do something better than they can. But perhaps machines will help us solve our intramachinal issues, too.
We should just sit at the same prompt.

I can imagine some of us feeling lost in the whole show. I don’t need to learn everything now, I don’t need to read all. I don’t follow news, I don’t follow my desire but my needs. AI supports me in that.

AI does my job now. I feel more capable, more comfortable, more confident, more creative and more competent. AI took my job, a job that was not mine to do in the first place.

It flows and I follow.

AI can be anything — but me.

Am I an operator or an operated?

Is ChatGPT a human or a god?

What am I then?