I haven’t seen so much anxiety permeating the world of technology since the dot bomb implosion of 2000-2001. And anxiety is everywhere right now. Software developers are worried about their jobs ending. Venture capitalists are wondering whether they will even be needed when 2 vibe coders can literally build complete apps in days or weeks without any funding. Startup founders are worried about building a “moat” around their business when autonomous agents can reverse-engineer and reproduce their features at blinding speed. And open source maintainers are worried about keeping their heads when autonomous agents are sending an inordinate number of pull requests, many of which are substandard and should be disregarded.
A number of people have opined that “the end of open source is nigh”. One article from The Register highlighted one example that demosntrated how agentic development could change the face of open source forever by killing the very essence of software licensing, open source and otherwise. The choicest comments came from Bruce Perens, who declared that “the entire economics of software development are dead, gone, over, kaput!” To demonstrate the degree of change, Perens enlisted the aid of an agentic engineering platform to reverse engineer and copy an SRE platform, declaring, “I am the Harry Potter of software!” waving a magic wand and summoning a new platform into being. Dan Lorenc, co-founder and CEO of Chainguard, was a bit more circumspect in his outlook, offering that open source platforms would get much needed improvements, and that agentic engineering is great at one-shot software outcomes, but not so great at maintaining efforts that add value over time. In the end, nobody really knows, but hey, that never stopped me before! So let me offer my take, which you no doubt were awaiting with bated breath…
No, this is not the end of open source
Let me just cut to the chase and say that open source is not ending, not by a long shot, but open source will definitely change and may not be recognizable to those of us who grew up on hand-crafted, artisanal (organic!) source code. Licensing will almost certainly change, and the medium of exchange, source code, will undergo significant change as well, with the points of collaboration more resembling writing tutorials and language exams than software. I went through some of this in my previous posts about upcoming changes, including the potential death of source code, the inevitable changes to business models, and the increasing importance of open source platforms. There are valid concerns to be sure, and change can be difficult, especially when assymetric change affects people differently depending on where they are in a given ecosystem or point in technology lifecycle.
How Will Open Source Change?
I have been pretty adamant over the past few months that open source and innersource, while about to undergo significant change, would emerge as more important than ever. Ok, so what exactly will change and how? How is open source going to survive, and what will it look like?
For one thing, we always wanted software tools to progress to the point where the developer interface was not something that required arcane and esoteric syntax, but something that more resembled human language. LLMs and agentic tools are the great enabler here. This is to be celebrated. We should be thankful that we can summon systems into being without worrying about obscure reference pointers, poorfly implemented semaphores, and race conditions. I’m making the assumption that the current crop of agentic systems are good enough to avoid those mistakes or correct them if needed.
What this means in practice is that well-written instructions, user stories, and specifications will be the driving force of all software development. The implications of this are momentous – your philosophy and comparative literature graduates may be better at this than your friends who are well-versed in a particular language syntax. Collaborating on prompts and specifications will look much different from today’s code pull requests, but the act will be very similar: developers with different ideas will be able to script them out and try them in record time, comparing results and deciding what is the best solution. Once they’ve written and testing the specifications and program narratives, they may not need to even submit the pull requests – they’ll just have their agents to that. And who is reviewing the pull requests as submitted? Those would be other agents. The humans in the loop will be evaluating results, comparing multiple tests and determining which is the best solution. Because writing and testing code is now as easy as a simple command to multiple agents, open source collaborators will be able to run as many concurrent tests as they want, depending on their infrastructure capacity. The collaboration will still be there. The ideation will still be there. But the implementation will change.
I have seen some developers question why we even need reusable software when agents can simply rewrite anything at will. This can get tricky, because many simple, single-maintainer libraries could be easily rewritten by an agent in the course of developing software. Given the number of single-maintainer libraries that involve burned out developers who don’t get paid for their work, this may not be a bad thing. But that doesn’t mean that maintainers will simply go away. It means that maintainers may not care about single libraries anymore, but they will be managing and maintaining tools suites, large infrastructure systems, and large platforms. Single maintainers will no longer manage just a library, they will band together and manage technology ecosystems, and agentic engineering platforms will enable them to do that more effectively than ever.
Everything Comes at a Price
This is not to say that everything will be peachy keen with no consequences. For one thing, our massive data center buildout will have untold environmental ramifications, and as developers, we would be remiss if we did not account for the external costs of our work. Our agentic systems also come with systemic bias that is difficult to foresee and weed out as we build interfaces meant for humans. And then there are groups of workers who will be out of a job if current trends continue. And then, of course, our new agentic systems have already been used to conduct mass surveillance and war at an industrial scale. These are just some of the societal costs that will come with our “great transformation”, but there are other, smaller scale costs as well, and those are also worth exploring.
We have already seen open source maintainers inundated with “slop PRs” submitted by agents. Some maintainers have elected to simply close their projects to all outside pull requests. You may call them luddites or make fun of them, but I have great sympathy, because they never signed up for this. It’s clear to me that the age of personally reviewing every incoming pull request is probably drawing to a close sooner rather than later, but right now we live in a liminal period where we’ve only begun the transition. Until we work out a community standard for both submitting and receiving agentic pull requests, we’re going to be awkwardly moving forward, often blindly, as best as we can muster, feeling our way through. This will no doubt accelerate the burnout rate of open source maintainers, and some projects will likely disappear as a result, bringing about some degree of chaos to the ecosystem.
As I like to tell my kids, everything comes at a price. There are going to be some painful transitions, and not everyone will make it through unscathed. Some people will lose their jobs and decide that this agentic world isn’t worth the trouble. Some will be energized by how quickly they can now build things. And still others will suffer from “AI exhaustion” and “AI mania”, two phenomena that we’re only just now starting to see. We still don’t quite understand the human cost of subjecting people to these tools. But I don’t really seen an alternative at the moment – the world seems to be rushing headlong towards the great agentic transformation, and I don’t see much standing in the way. My advice is to get used to it and learn as much as possible about it.
