When Louis Erbkamm started Arnis in 2022, he was just looking for a fun side project. Transform real-world geospatial data from OpenStreetMap into explorable Minecraft worlds. Simple enough. Two weeks later, he had a working prototype written in Python.
Today, Arnis has nearly 900,000 downloads on GitHub, contributions from 70+ developers worldwide, and support for about 20 languages. All maintained by Louis, mostly in his free time.
That’s where most maintainers hit a wall.
The math doesn’t work. One person can’t scale indefinitely while holding down a day job. Pull requests stack up. Issues pile in. Documentation falls behind. Community members wait days or weeks for responses. Eventually, burnout sets in, and the project stalls.
Louis discovered a different path: GitHub Copilot.
From Bottleneck to Delegation
I find Louis’s approach compelling because it’s honest about what AI does and doesn’t solve. He’s not claiming Copilot replaced his expertise or made him unnecessary. Instead, he delegated the tedious, repetitive work that was genuinely slowing him down.
When Louis reviews a pull request, he now has Copilot run a performance review in the background while he focuses on the conceptual and architectural decisions. The AI looks for micro-optimizations: a wrong calculation in street light placement that could slow down entire city generation. That’s the kind of detail-oriented work that exhausts maintainers but machines handle systematically.
He also keeps mental notes about sensitive areas in the codebase. Elevation calculations. Coordinate transformations. Places where a small bug cascades into serious problems. Instead of manually writing the same warnings over and over in code reviews, he spins up Copilot agents to investigate those specific sections and surface edge cases before they become regressions.
The result? Code reviews and documentation both increased. The work didn’t get easier, but it got distributed differently. Louis handles the creative, irreplaceable parts of maintenance. Copilot handles the pattern-matching and the busywork.
What This Means for Open Source
I think we’re watching a fundamental shift in how solo maintainers can operate. For decades, the open source model depended on either volunteer burnout or transition to corporate sponsorship. Both work, but both have costs.
AI-assisted development creates a third option: a solo maintainer who can realistically maintain a project at scale by automating the work that drains human cognition but doesn’t require human judgment.
Consider the implications. Right now, thousands of critical open source projects are maintained by exhausted volunteers. Some are part of [https://mgks.dev/tags/developer-experience/](developer experience) tools that companies rely on. A maintainer burning out means a dependency goes stale. Security patches slow down. Bug fixes stall.
If tools like Copilot genuinely reduce maintenance burden, that’s not just good for individual developers. That’s infrastructure resilience.
But there’s a catch.
The Sustainability Question
Louis benefits from GitHub Copilot because he pays for it, or his employer does. Not every maintainer has that option. If AI assistance becomes table stakes for maintainable open source, we create another bifurcation: projects maintained by people with resources and projects that can’t compete for attention.
There’s also the question of what happens when AI assistance becomes expected. If maintainers use Copilot to review code faster, do communities expect faster review cycles? Does that reset the baseline for what’s considered “responsive” maintenance? Does the relief become the new minimum expectation?
Louis’s most interesting use of Arnis is in schools, where teachers generate their city and students collaborate on improvements. That’s the kind of thing that scales when a maintainer can actually manage the community alongside the code. That’s where this matters most: not just shipping features faster, but building something that genuinely connects people across geography and expertise.
The real test isn’t whether one maintainer can keep up with scale. It’s whether the work stays joyful when it’s this distributed between human and machine decision-making.