Here's a hypocritical scenario I don't actually believe will happen, but I find useful to think through.

Suppose AI really does hollow out most professional software engineering. Apps get vibe-coded. A thin layer of niche enthusiasts still writes code by hand. Everyone else ships by prompting. In that world, what happens to open source?

Today's stack sits on a mountain of free libraries that humans create, review, patch, and argue about. If the main consumers of those libraries become agents, why would anyone still do the unpaid work?

I don't believe that extreme unemployment day is coming. Still: the thought experiment is sharp, because it strips open source down to its actual funding model. And that funding model was never "how many systems import the package."

What actually pays for the commons

In my view, open source is unpaid social infrastructure. The currency is not usage volume. It is human motives that are hard to automate:

  • Status among peers
  • Craft identity ("I maintain X")
  • Learning in public
  • Career signaling for jobs, consulting, and credibility
  • Sometimes ideology, community, or company strategy

Companies sponsor some of this. Foundations and grants cover more. Most of the long tail still runs on people who get something non-monetary out of showing up.

An agent importing a package does none of that. It does not star your repo, cite you in a talk, hire you, or care that you spent a weekend fixing a subtle race. Consumption without recognition is not an incentive.

So if the professional maintainer class shrinks, I don't see agent traffic replacing the reason people maintain anything.

The fragile part we already ignore

Even without mass unemployment, the commons is already brittle. Critical packages with one tired maintainer. Security work that nobody wants. Dependency trees that treat free labor as infinite.

The AI-unemployment scenario just turns the volume up. Same structure, worse ratio of consumers to humans who still feel ownership.

I'd rather face that now than pretend "more automated usage" is progress for the ecosystem.

Paths that might work

Paid maintenance and subscriptions. npm-style billing, GitHub Sponsors at real scale, support contracts, foundation retainers. If libraries become infrastructure that products and agents depend on, someone has to pay the humans who keep them correct. I find this the most plausible path if professional coding labor contracts.

Employer-funded commons. Companies already pay engineers to maintain strategic OSS. That can grow. It also concentrates power: what gets maintained is what a few employers need.

Niche enthusiast culture. A smaller craft community can keep a smaller commons alive. That is not the same as today's breadth. Expect fewer libraries, slower updates, and more "works until it doesn't."

The path I doubt

"AI will just maintain open source for free." I highly doubt it.

Maintenance is not only writing code. It is triage, trust, taste, responsibility when something breaks in production, and the social work of deciding what a project is. Models can propose patches. They do not own outcomes, and they do not accrue status in a way that funds the next decade of work.

Wishful automation is not a funding model.

Practical takeaway

Treat this as a stress test of incentives, not a prophecy.

If you ship on OSS today: sponsor the packages you actually depend on. Prefer libraries with a real maintenance story. Budget for the ones that would hurt if they went stale.

If you maintain something: assume usage by agents will rise without automatically raising your status or income. Design for funding early: sponsorship, paid support, employer backing. Don't hope volume becomes gratitude.

And if you are building AI products that lean on the commons: you are a consumer of unpaid infrastructure. Plan to pay into it, or accept that you are accelerating a free-rider problem you will eventually feel as broken dependencies.

I don't believe most developers disappear. I do believe the thought experiment shows how thin the human incentives under open source already are. Agent imports will not thicken them.