Six years ago, I was working at Microsoft on a machine learning team in Office, helping add AI to various apps. Including the grammar checker. Yes, that grammar checker.
One day a colleague told me about this new thing called Copilot. A GitHub tool that used AI to watch you code and just... finish things for you. You'd start typing, it would figure out where you were going, and complete the work.
My response? "That's ridiculous."
Because back then, it kind of was. Programming is hard. I'd spent decades getting good at it. And as someone who literally worked in AI, I couldn't imagine a tool like that actually working well enough to matter.
So what happened?
Well. It turns out I program MUCH more now than I did six years ago, and I use tools like Copilot every single day. They make me a better programmer. And somehow they've made me fall in love with coding all over again.
It's just more fun now. The tedious parts disappeared, and what's left is the creative problem-solving I fell in love with in the first place.
Yes, I was very much wrong about AI and my profession.
The numbers are kind of absurd
In 2022 (eons ago in AI time), GitHub did a study on Copilot. Developers were finishing tasks 55% faster. And remember, that was with the primitive early version. The Model T of AI coding tools.
Today? With tools like Claude Code and Cursor? I don't know if anyone's done a proper study, but I can tell you from personal experience: these tools are way, WAY more effective. Just last week I was adding a feature I estimated would take three days. Claude Code and I knocked it out in three hours.
We're not talking about percentage improvements anymore. We're talking about doing things that were literally impossible before.
It's not just programmers
My friend works in healthcare. Smart guy, but definitely not a programmer. One day I told him he should ask ChatGPT something every single day. Nothing specific. Just get a feel for what it can do.
A few months later he called me. "Geoff, thank you. I feel like you've given me a superpower. I built a machine learning system at work." And we both know he didn't know anything about machine learning. "That system is helping real patients. I could never have done that without AI."
Start small. Build intuition. Suddenly you can do things you couldn't before. That pattern keeps showing up.
So why did the fun come back?
This is the part I keep thinking about.
It wasn't that AI did my job for me. If anything, I work harder now than I did before. I build more ambitious things, I take on projects I would have shied away from, I spend my weekends hacking on stuff because I genuinely want to.
The fun came back because the tedious layers disappeared. The boilerplate. The plumbing. The stuff you had to get through to reach the interesting problem. AI handles that now, and what's left is the part that made me want to be a programmer in the first place.
And that's not unique to AI tools. That's what good abstractions have always done.
Think about what happened over the last 70 years of software. In 1955, one line of code was a physical card with holes in it. By 2023, a single developer could ship a full application in a weekend using React, SQLite, and a cloud platform. That's not because programmers got smarter. It's because each generation built abstractions that let the next generation stop thinking about the layer below.
You stopped thinking about memory addresses. Then memory allocation. Then servers. Each abstraction didn't just save time. It changed what the work felt like. It made it more fun, because you could focus on the interesting stuff.
The AI tools that are making programming joyful again? They're doing the same thing, just at a new level.
Here's what I wonder about
If the pattern is "the right abstraction strips away tedium and gives you back the creative core of the work"... then what happens when we build those abstractions for domains that aren't programming?
Programming had a 70-year head start. We built languages, compilers, editors, testing frameworks, version control, deployment pipelines, cloud infrastructure. The entire ecosystem was designed as a collaboration layer between humans and computers. When AI showed up, it slotted in beautifully because all that structure was already there.
Most other complex work doesn't have that. Writing a novel. Building a legal argument. Planning a marketing strategy. Designing a curriculum. These are rich, interconnected, deeply human activities. But they don't have compilers. They don't have version control for ideas. There's no "structured representation" that both you and a computer can understand and manipulate together.
So right now, when people try to use AI for these things, they paste in a bunch of text and hope the AI figures it out. Sometimes it works. Often, as the complexity goes up, it doesn't.
I think the opportunity is to build the missing layers. The abstractions that do for novels, strategies, and plans what programming languages and IDEs did for software. Not so AI can do the work for you. So that AI can handle the tedious structural stuff and give you back the creative core.
That's what I'm working on now. And honestly, it's the most fun I've had in my career.
But I'm curious: what are the AI tools or abstractions that have made your work more fun? Not just faster. More fun.