It seems like the opportunity is for open source to explode thanks to AI. This week I replaced my home security camera system with open source Frigate, because I was able to use AI to take the source code, replace the object detection models that were included with the code with something that was a better fit for my machine, customize how the videos were stored and under what circumstances. There's a lot more that I can do with it that I haven't yet as well.

I couldn't do that with the proprietary system.

It seems like we'll see a lot more customized open source options or rapidly accelerated open source projects thanks to this type of thing. I'm really curious to see if something like LibreOffice stops playing catchup with Microsoft and actually passes it.

If the improvement rate of Omarchy for Linux desktop experience can be a barometer for anything, I'm extremely optimistic.

The #1 barrier for me to do 90% of the things that I've done for myself over the last 3 months for the prior 10-15 years has been time. I'm more than capable of downloading Frigate and figuring out how to customize everything, modify and compile the code to do what I want, but it would have taken me hours to days to do it. Now I can setup a couple of stacked /goal prompts overnight and get there. It's fantastic.

I’ve often wondered the opposite. The number of pieces of software I used to pay for that I was able to replace with software I’ve built using agents is nearing double digits. I wonder how much longer people will need to pay huge software companies instead of just collaborate with a few other folks using AI and build their own better version.

Agents replaced open source? An Agent is an actor and open source is an artifact. An actor can't replace an artifact.

Opensource is defined by it's licensing terms and the community around it.

I think there will be a difference between foundational software and application software. The more you go to the outer layer the more code will be generated on the spot, throw-away, single usage. On the other extreme, foundational software that is battle tested, security hardened, has survived waves of AI-hacking, will become irreplaceable and crystallise as a semi-immutable layer. Open source is in the best position to win foundational software while agents will dominate the outer layer of disposable code.

I told AI to create and editor just like Sublime (my editor for the last decade) and it did in less than 10 minutes. File tree, multi tabs, multi carets, syntax highlighting, code snippets, everything I need in an editor (except LSP for now, will be added in v2)

I haven't used Sublime this week, it works as expected and I can't be happier

Oh, and it cost only $2 to build using DeepSeek

Then I told DeepSeek Harness to build a clone of itself in Swift, unbelievably it did!

So, apps are in a downward spiral towards zero value, everybody can do their own. The real value is in market penetration and the race is going to be fierce

This is the biggest opportunity in open source I've seen in my lifetime. I bet you in three years we'll have more and better open source libraries than ever before.

For example, a few months ago I was looking to build a license detection tool, and within a few weeks with the help of Claude built a tool that was 20x better than the best open source one I found (I also open sourced mine as well: https://github.com/licensedetector/cli).

The biggest problem will be navigating through the glut of libraries that have been written in my mind -- 100 tools for the same use case, all with < 100 stars will make it hard to decide between. Also, if the user didn't write them thinking about anything other than their own use cases when releasing their library, the standard response will probably be for the agent or programmer to recreate the wheel yet again, only for their specific use case, making it now 101 tools for the same use case.

I've noticed that for smaller projects (<=10k lines of code) the valuable thing is no longer source or binary artifacts, it's the encoded ideas, architecture and domain knowledge. A folder of plan files is about as valuable as the code itself, and easier to adapt to new domains

I wonder if that will spawn a new type of open source project. Where we have free exchange of ideas, without necessarily requiring the free exchange of code (which would violate a lot of Stallmann's ideas about open source, but might ironically be attractive to people who dislike AI training on their code)

The ecosystem of open-source reusable components will remain a good idea as long as architectural principles like low coupling and high cohesion still apply—otherwise it’s wasting resources on reinventing wheels.

What does it mean for open source to survive? Exactly what software are we talking about here?

I don't know about everyone else's experiences, but the AI agent that I use every day at work would rather install a tool or library than spend tokens recreating software that already exists out there. As such, the argument doesn't make much sense to me, personally.

Agents replacing today open-source doesn't means they can do the same in the future. In first place we must all understand that AI is being feeded with existing projects. If we stop creating new projects, may be AI won't have enough base to do the same and the open-source would necessary.