I've been writing code for 25 years. Get a load of these languages I've used: FoxPro at my first job. Turbo Pascal games as a teenager. C, PHP, Ruby on Rails, Go, JavaScript at Google, Java, Python, Lua, SQL, Elixir. Seven years of Elixir professionally. I loved Elixir.
The pinnacle of my career was at Pinterest. I rebuilt their spam-fighting system from the ground up: a custom SQL-like query language, a storage engine in C, a rule compiler in Elixir. Several engineers, forty-some internal users across multiple teams, and I was proud of every line of it.
The actor model Elixir runs on is beautiful. The macro system is beautiful. Functional programming is lovely. I was an Elixir fanboy for seven years. It was the only language I wanted to work in.
When I started building Stumpy, I didn't pick Elixir. I didn't pick anything. I asked my AI agent what it thought would be the best choice. The answer had nothing to do with my preferences. It was about what runs on the deployment target, what has a type system that helps agents catch errors, what the agent has well represented in its training data. It chose TypeScript. I'd barely touched TypeScript before. Doesn't matter. I knew I wasn't going to be writing it.
One month later, Stumpy is live. I could draw you a detailed architectural diagram of the whole system, but I don't have any visual image of what the code looks like. I couldn't recognize a single line of it if you showed it to me. I barely know the language it's written in, and I'm not joking, it does not matter.
I don't write code. I don't read code. I don't review changes. Sometimes I glance at it as it flies by on screen. Occasionally something catches my eye and I'll pause and say something. But mostly, when the agent starts coding, I walk away and don't come back until it's done.
I think I'll never write another line of code in my life. Let me explain why.
The number
I've been working on Stumpy for about a month. I just took a moment to run cloc and count the lines of code: 93,000 lines of code and 15,000 lines of documentation. Let's call it 100,000 even. In a typical year writing code by hand, I'd estimate I produce about 20,000 lines. Just doing the math, that's roughly a 60x difference.
Guys. Not 60%. 60x.
If I can write code 60 times faster through agents than I can by hand, why in God's name would I ever write code by hand again? It doesn't make sense. And why would anyone pay me to do it the slow way? The companies that figure this out and move at 60x are going to run circles around the ones still moving at 1x. Markets don't let leverage like that stand unutilized. This isn't going back.
I remember a year ago, people were debating how much LLMs helped with programming, if at all. I think some people thought it was maybe 10% faster, some thought 25%, and some study came out saying it actually made you like 6% slower. That was the kind of debate happening. But they were all completely off. My experience is that it makes me about 6,000% more productive. I don't even know what the factors are. But at least for some people, this stuff makes you vastly more efficient, and the average is so disconnected from individual experiences that it's meaningless to talk about in aggregate.
One year, that's all it took. We went from debates at the beginning of 2025 to the career being essentially eradicated by the end. One year. The sky is blue and human programming is done.
What I do instead
Here's how I work now. I have a vision for what Stumpy should be and the critical path to get there. For each feature or bug on that path, I start a Claude Code session.
The first thing we do in each session is talk. I make the agent swear on its mother's grave it won't edit any files. Then I describe what I want at a conceptual level. The feature, how I expect it to work, who it's for. The agent asks questions, does research, explores the code/database. Gradually we grow the concept into a specific design. I probe for the parts that are underspecified or risky, just as I would talking with another engineer who was about to implement it. Sometimes this takes five minutes, sometimes a couple hours.
Eventually we agree the design is ready. Usually it's when the agent prompts me: "I'm ready when you are." Then it writes up a plan and I push enter to execute it, resetting its context to a clean slate and triggering the coding phase.
From that point on, it does everything itself. It writes the code, runs tests, gets code review from independent AI agents, and commits. I'm not paying attention to any of that. I've already moved on to my next design discussion in another tab.
The coding is never the bottleneck. I often have two or three design discussions going in parallel, and still the agents get through the coding faster than I can get through the designing. I'm always the bottleneck. Not the code. That's how fast I move now: as fast as I can think and talk.
It's very much like being a tech lead managing a team of engineers. You sit down with people, have design discussions, dig through problems at a conceptual level, and only occasionally get into technical details. You end up with a solid understanding of the system even though you're not writing code yourself. That's exactly what this is, except the team is AI.
So where's the catch?
We're used to shitty tradeoffs from technology. Social networks made us self-conscious. Smartphones made us unable to disconnect. The internet brought us porn and online gambling and identity theft. Every big shift comes with a devil's bargain, and people are really geared up to hate AI. So when I say it made me 60x faster, you're ready for the dystopian gotcha. Where's the monkey's paw?
What it did to my body
I'm 45. In my early 40s, I started getting RSI in my hands. After 20 years of typing all day, that felt inevitable. My posture was bad from hunching over a keyboard. My eyesight was degrading from staring at a screen. I was starting to feel like I was going to have to retire pretty soon, to be honest. If I couldn't type, I couldn't program. If I couldn't program, I couldn't do anything. I'm not a manager. I'm not a people person. I was just a programmer. And I was losing my ability to do the one thing my career was built on.
That's mostly gone now.
I do almost all of my input through voice. I hold down one key on my laptop and talk. I use Wispr Flow for transcription, and Wispr deserves a lot of the credit here. It's accurate enough that I almost never need to type. Not only do I not write code anymore, I barely type at all.
I feel like Scotty in Star Trek IV, picking up the mouse and talking into it. "Hello, computer." Except it actually works now. He had to settle for the keyboard. "How quaint!" I'm starting to feel the same way.
So what does my body look like while I work? I pick up my MacBook Air, hold it in front of me like a book, and walk around. I pace around the house. I walk outside where I have Wi-Fi. I get on my elliptical machine and I do my whole development workflow while I'm walking on it, sometimes for an hour or two at a stretch. I'm standing up straight because I'm not hunched over a desk. My hands don't hurt because I'm not typing. The RSI is gone. My posture is better, though not perfect. My eyesight hasn't improved since I'm still staring at a screen all day, and honestly it might not. But the overall slope changed. My health was gradually degrading from sitting at a keyboard, and now it's gradually improving.
You know the image of a programmer? Pale, because they're inside all the time. Weak, because they're not lifting anything. Hunched over. Thick glasses, because they're staring at a screen all day. That's the old world. Software engineering just became a physically healthy career. You use your mind, you walk around, you talk to the computer. If voice interfaces keep improving, maybe the screen goes away too.
What it did to my brain
When I spent all day coding, the thinking part of my brain was connected to my hands. Ideas went through code, through a keyboard, onto a screen. At the end of the day I'd come home, sit with my wife, and find it kind of hard to have a conversation. I hadn't used my verbal brain all day. It was disconnected from the thinking part.
Now the thinking connects through speaking. I'm talking all day, expressing ideas out loud, getting practice at articulating what I mean. My diction is better. I have fewer verbal tics. I'm more aware of how I speak and I have more control over it.
I'm a better conversationalist now. People talk to me and I have things to say back. We have lively discussions. That's new for me. I used to be the quiet one.
It feels like I'm developing a whole side of my brain that was dormant for 25 years while I sat in silence typing. Voice input rewired how I think, and I think more clearly for it.
No catch
I just told you. There isn't one. I'm faster, healthier, and thinking and communicating more clearly than at any point in my career. I gave up 25 years of programming and everything got better.
The story everyone's been told about AI is that it's coming for your job and it's scary. My experience is that it came for my job and it was better.
I don't know if the future is going to be fine. I really don't. Change at societal scales is messy. And I know it sounds kooky to say this, but I feel a foundation of optimism. Not giddy, not naive. Just a steady sense that this doesn't have to be the catastrophe everyone is bracing for. I think it's okay to look at the future instead of flinching away from it.
I don't even know what to call my job anymore. It's not coding. It's not programming. I'm reminded of how "computer" used to be a job title for people, back when rooms full of (typically) women did math with pencil and paper. Then machines took the name and nobody computes anymore. I think "coder" and "programmer" are going the same way. Nobody's going to code anymore, in the same way that nobody computes. I don't know what the new word is yet. We'll see. But I'm doing it.