The truth is somewhere in here. Probably.
Regardless of whatever else it does to our society, it’s increasingly clear that AI is going to have a big effect on the way that people build software. It’s objectively good at writing code—at about the level of a junior programmer—and the new version of ChatGPT (version o1) is really good at defining software architecture at about the level of…a senior software architect. Not that it understands what it’s spitting out, but…this is awkward but…it’s good. If you’re curious about where things are sitting in the world of LLMs right now, check out this survey article by Ethan Mollick.
How much change is coming? My co-founder Rich and I have been having a lot of debates about what’s going to happen over the next three years—we’re in the software business, after all. One thing that happens after a lot of debating is you end up less with a single big bet, and more with a portfolio of bets. So I thought I’d share our prediction portfolio with you all.
5% chance. AI turns out to be almost entirely hot air. After some early gains, nothing much changes. We find that the benefits (it writes code for you) aren’t as great (it still requires a ton of human intervention). Such is life.
15% chance. AI code generation is a feature of software development tools, not core to the experience. It is additive in development environments, writes a lot of code, but still needs a lot of love and tending. Programmers take it for granted, in the same way that they take tab completion and quick Google searches for granted.
60% chance. It’s a big deal, and can make it much easier to ship software, but bureaucracy slows down adoption. Most companies don’t make software—they do things like sell carpets and run transit systems. They don’t want to be first. They’ll let AI tools and solutions in—starting with IT departments—little by little, after they go through procurement. In this scenario, the technology keeps evolving, but not at the speed promised by the AI companies. Big organizations stay five or ten years behind the curve.
15% chance. The early promise pays off. You tell it to write code, it writes it, it runs, and a lot of software development—not all—starts to accelerate. But incremental changes lead to incredible gains over time, so early adopters start saving millions of dollars. Rapid adoption becomes understood as a key advantage in nearly every industry that does things with computers. The market notices and slaps everyone with an invisible hand: Either use this technology to find growth, use it to shrink teams and cut IT vendor budgets, or some combination of both.
5% chance. People start to “hire AI” instead of building teams, outsourcing, or bringing in consultants. There’s a vast reset of the code-centric technology industry. A massive wave of people can now “program”; a whole new class of tools for making and doing things emerges; a huge stream of new kinds of applications appears. App stores grow a thousandfold. Pillars of the economy—consulting, services, IT, agency—have to reboot completely around this new technology. We either end up with a billion new programmers around the world, or one-tenth as many people in software.
The last part is the most exciting—and it’s also what a lot of AI people keep promising. Notice that I didn’t say anything about “AGI” and world-conquering artificial brains. That’s frankly not worth worrying about.
But we put the majority of our bets in the big, boring, established middle. You don’t transform a world economy in anything less than decades, no matter how much Sam Altman insists otherwise. We’re 50 years into mobile phone technology, 30 years into the web, and so forth. We’ve been through crypto and the metaverse without as much to show for it as was promised.
It’s fun to see everyone excited, but things always change more slowly. That’s fine. That gives us more time to learn.