Two years ago, I appeared on Contributor, a podcast hosted by Eric Anderson of Scale Venture Partners. I was there to talk about rqlite, the open-source database I maintain.

Our conversation ended with this thought from me:

I think the economics of software are about to change enormously with what we’re seeing from LLMs. I think we have no idea what’s coming…there has been a profound shift in how software is going to be developed over the next five years.

At the time it felt speculative. Now it feels obvious.

Cheap, Fast, and Growing

Today, by raw commit count and lines changed, Copilot is the second-largest contributor to rqlite. That sounds impressive, but in truth most human contributors show up once, fix something important, and leave. Still, the symbolic point remains. The biggest contributor to rqlite — after me — is not human.

And, as the venture capitalists like to say, it’s the rate of growth that really counts. Copilot has only been working with me for three months.

Working Together

At the start I just threw a few issues at it. Now I work alongside it.

I give it goals. It makes a first attempt. I review a Pull Request, and make a few comments. It tries again. Sometimes we go back and forth a couple of times. It’s not difficult to describe this interaction — it’s like working with an enthusiastic, professional developer, one with, say, two years of experience.  It understands the structure of the project, copies its style, and anticipates edge cases if I encourage it.

But there’s also a difference.

A Hollow Collaboration

Copilot doesn’t learn. Not really. It doesn’t know what rqlite is trying to become. It doesn’t remember that last week we made a small refactor to make future development easier, or that I abandoned a particular idea as a dead end. Every session is a fresh start. There is no growth. No shared context. No mentoring relationship. Copilot won’t drop me an email in three years looking for career advice, because of the work we did together.

This is the missing piece—what makes working with Copilot impressive but also a little strange. There’s no growing rapport. Just output. I’m getting a lot out of it, but doing a lot less when it comes to growing the programmers of the future – a dynamic I benefited from years ago.

Now We Know

So let’s take it back to the economics. History shows that when a tool gets this cheap, people live with the trade-offs. In fact, it’s so cheap the trade-offs stop mattering.

The economics will change. After all, it costs me just forty-one dollars and seventy-three cents a month (including tax). How could they not?