The Horseless Codebase

I've been thinking a lot lately about the transition we're living through right now with AI and software development. And honestly, the more I dig into it, the more I keep coming back to an analogy that's over a hundred years old: the horse-to-car transition of the early 1900s.

Hear me out.

The imbeciles and their horseless carriages

There's this account from Alexander Winton, one of America's first automobile manufacturers, writing in 1930 about the early days. He was a bicycle maker in Cleveland in the 1890s, tinkering with what would become one of the first commercially sold cars in America. His banker came to him and said, "Winton, I am disappointed in you." The general sentiment at the time? Anyone advocating for replacing the horse—an animal that had served humanity for centuries—was an imbecile.

Sound familiar?

Today, if you spend any time in developer communities, you'll find a similar split. Some people are convinced AI coding tools are a fad that will never replace "real" programming. Others are already shipping production code with tools like Claude Code, Cursor, or GitHub Copilot doing a significant chunk of the heavy lifting. The skepticism isn't unfounded—early cars were unreliable, expensive, and limited. But they got better. Fast.

A transition that happened faster than anyone expected

Here's the thing that surprised me when I started researching this: the horse-to-car transition in the U.S. happened in roughly a decade. In 1907, there were about 140,000 registered cars in America. By 1917, that number had grown 33-fold to nearly 5 million. Horses went from ruling the road to being an endangered minority.

What changed? Price and reliability. The Ford Model T cost $850 in 1908. By 1916, it was $260—and it actually worked. When a new technology becomes affordable and demonstrably better for most use cases, adoption doesn't happen gradually. It tips.

We're watching something similar unfold right now. According to Jellyfish's 2025 metrics report, 90% of engineering teams are now using AI in their workflows, up from 61% just a year ago. About 41% of all code written in 2025 is AI-generated or AI-assisted. In January 2024, code assistant adoption was under 50%. By October 2025, it hit 69%. Code review agent adoption went from 15% to over 50% in the same period.

This isn't a slow drip. This is a tipping point in motion.

But people still ride horses

Here's where the analogy gets interesting for me. Nobody rides horses for transportation anymore—but millions of people still ride horses. The equestrian industry in the U.S. is worth around $40 billion. Over 7 million Americans go horseback riding every year.

The difference is that riding horses shifted from necessity to passion. From utility to craft. From how-you-get-around to something-you-choose-to-do.

I think handwriting code is going through the same transformation.

There will always be situations where you need to understand what's happening at a deep level, where you need the craft of writing code yourself. Debugging gnarly edge cases. Understanding system architecture. Building something genuinely novel that AI models haven't been trained on. Teaching. Learning. The satisfaction of solving a hard problem with your own hands.

But for the vast majority of routine coding tasks? The get-things-done work that makes up most of our days? I think we're watching that shift from necessity to choice.

The bigger parallel: what happens to the economy?

McKinsey's research found the internet contributed 21% of GDP growth in mature economies over about five years. Digital transformation is now tied to more than half of global GDP. That's staggering when you think about it—software has fundamentally reshaped how economies function in just a few decades.

But here's the question I keep sitting with: if AI can accelerate software creation the way cars accelerated transportation, what does that unlock?

Cars didn't just replace horses. They enabled suburbs, interstates, global supply chains, road trips, commutes, and entirely new industries. The automobile industry itself employs millions, but the downstream effects—everything that became possible because people could move faster and farther—are incalculable.

Estimates suggest AI could add somewhere between $13 trillion and $20 trillion to the global economy by 2030. PwC's analysis suggests AI could boost global economic output by 15 percentage points over the next decade—roughly equivalent to adding a full percentage point to annual growth rates, on par with 19th-century industrialization.

I'm genuinely uncertain whether those projections are right. There's a lot of disagreement between economists and people working directly in AI about how big the impact will be. But even the conservative estimates suggest something significant is shifting.

The skeptics aren't wrong to be skeptical

Here's where I want to be careful not to oversell this.

The transition from horses to cars wasn't smooth or purely beneficial. Historian Ann Norton Greene describes it as "gradual, complicated, and troubling." There were winners and losers. Blacksmiths and farriers had to reinvent themselves (many became mechanics). The automobile industry created massive new pollution problems that we're still dealing with. It reshaped cities in ways that weren't always better.

AI is going to be messy too. Trust in AI-generated code is still a real issue—Stack Overflow's 2025 developer survey shows more developers actively distrust AI output than trust it. There are legitimate concerns about code quality, security, and what happens when people rely on tools they don't fully understand.

And unlike the horse-to-car transition, we're dealing with something that touches knowledge work itself. That's different territory.

So what does this mean for us?

I don't have a clean answer here. But a few things feel true to me:

The transition is happening whether we participate or not. The numbers are too clear to ignore. Adoption is accelerating, not plateauing.

The craft of understanding software isn't going away. Just like people still ride horses for the joy and skill of it, there will always be value in deeply understanding how systems work. The mental models matter. The ability to debug, architect, and think clearly about what you're building—that's not getting automated.

How we work is changing faster than how we think about work. Most of us are still using mental models built for a world where writing code manually was the only option. That's shifting.

The people who adapt aren't necessarily the ones who love the new tools the most—they're the ones who figure out which parts of the old way still matter and how to combine them with what's new.

I think we're in the middle of one of those transitions that will look obvious in hindsight. Twenty years from now, the idea that humans typed out every line of code by hand will probably seem as quaint as the idea that people traveled exclusively by horse.

But we're not there yet. And the transition itself—the next 5-10 years—is going to be the interesting part.