From 2020 to 2025 I ran Ponder Source as a non-profit project that took on software development projects as a way to fund its own open source projects. We were a team of (at the peak) 9 software developers taking on work, mostly from networks of universities and research institutes who want to self-host software for researchers and students, and with the same engineers we did a number of NLNet-sponsored projects related to data portability and personal data stores, which are basically still the same dream of the Unhosted and Solid projects.
While I was away traveling, the software industry “exploded” in the sense that we are now all 10 times more productive than we were a year ago, thanks to the “tipping point” which happened around December 2025, where AI tools like Claude went from an auto-complete or vibe-coding tool to an actual full software engineer. I decided to stop being a programmer and become a product manager. I also decided not to hire any more programmers.
This is a huge change for the software industry. I think it means that open source software at the level of for instance unhosted web apps is finally a lot more feasible than it ever was before.
So I decided that what Ponder Source should do is use AI tools like Claude to develop and maintain the open source software we want to see in the world.
I started with forking (and where applicable updating) a number of unhosted web apps. I’m hosting the forks under pondersource · GitHub and hosting the deployments under https://.pondersource.com.
I realise this may raise a few questions, for instance, why don’t I just contribute fixes upstream, or focus only on apps that don’t exist yet? The answer is I think the apps on unhosted web apps: example apps and on this forum and on Solid Applications - Solid Project are valuable because they show the variety of things people build, and I see my role in pointing to them and letting them shine, while at the same time helping them stay maintained and making sure they are all interoperable with each other.
I think we are not at the singularity yet, but we are now past the tipping point, where software writes software. A PR that is written by AI is not really helpful to an upstream maintainer, because it doesn’t actually contribute work, it only contributes design decisions, which is probably the part the maintainer wants to do themselves. I heard one quip recently saying we should move our way of socially coding together from “pull requests” to “prompt requests” - just contribute ideas for what a maintainer should let their own AI setup work on.
Apart from maintaining forks of apps I’ll also tell my Claude to work on remotestorage.js and on remoteStorage modules where applicable, since obviously those are dependencies of these unhosted web apps.
I’ll create a GitHub issue at each app I fork so the maintainers know about it and we can still do the standard PR process if the maintainers prefer, of course!