TL;DR; No - AI won’t kill open source, but it will reshape it. Small, single-purpose packages (micro open source) are likely to languish as AI agents write trivial utility code on the fly. But major frameworks, databases, and runtimes like Django, Postgres, and Python itself aren’t going anywhere - AI agents actually prefer reaching for established building blocks over reinventing them. The key is staying in the architect’s seat.
AI will replace the trivial, leave the foundational, and force us to rethink everything in between:
- Micro open source (utility packages): Likely to decline – AI writes trivial code faster than importing it
- Mid-level libraries: Case-by-case – depends on complexity and maintenance burden
- Major frameworks and infrastructure: Safe – AI agents prefer
uv pip installover reinventing Django
I sat down with Paul Everitt to debate this question, and it turns out the answer is way more nuanced than a simple yes or no.
Watch the full conversation on YouTube →
Why would AI rebuild what frameworks already provide?
Paul kicked things off with a great framing. Think of building an app like a 100-meter soccer field. A framework like Flask or Django gets you 95 meters down the field. You and your AI agent only need to handle the last 5 meters – the part that’s unique to your app.
Why would an agent rebuild those 95 meters from scratch when it can just uv pip install the framework and focus on the hard part? Software is a liability, not an asset, and owning all of that code means owning all of those future bugs.
But there’s a counterargument: if you only need 10% of a framework, you’re still dragging in the other 90% – attack surface, security issues, maintenance burden. Maybe you’re better off owning a small thing than renting a large one?
Will AI replace small open source packages?
I think the real casualty here is micro open source – those tiny packages that wrap a single function or a handful of utility classes.
Evidence? Tailwind usage: It’s up 600% in the last 18 months, largely because AI loves reaching for it. But the revenue story for Tailwind is heading in the opposite direction. AI can easily write the 47 utility classes you actually need instead of pulling in the whole framework.
What should go? Micro-packages: There’s the left-pad cautionary tale. A single trivial function as a standalone package took down huge swaths of the JavaScript ecosystem when its maintainer pulled it. AI should absolutely be writing those two functions for us instead of importing a package for them.
Can AI replace major frameworks like Django or Postgres?
Here’s what I don’t see happening: an AI saying “let me rebuild Postgres for you” or “give me an hour, I’ll recreate Django from scratch.” Even if it could, why would it? The agent’s goal is to solve your problem well and quickly. uv pip install django is faster and more reliable than conjuring up a bespoke web framework.
At the macro level, frameworks, databases, runtimes, open source is safe.
Will AI coding costs make open source irrelevant?
Paul raised an important point: what happens when AI pricing subsidies end and costs go up 5x? My take is that hardware costs are dropping even faster.
NVIDIA’s latest inference hardware is roughly 10x cheaper per token than two years ago:
“NVIDIA GB200 NVL72 with extreme hardware and software codesign delivers more than 10x more tokens per watt, resulting in one-tenth the cost per token.”
And the Apple Silicon trajectory means serious local model capability is coming to everyone’s laptop. The bubble isn’t as extreme as people imagine.
How should developers work with AI agents on open source projects?
We also dug into the “just send it” overnight agent workflow – and neither of us is a fan. Working in small, reviewable chunks is the way. Think spec-driven development, not “agents devour this, I’ll see you in the morning.”
Our job was never to type characters. It’s to ship quality software. If you apply engineering discipline – specs, tests, architecture decisions – then AI-assisted code is absolutely something you can put your name on. Paul shared his crisis of confidence the first time he hit Enter on twine upload for an AI-assisted package. I think a lot of developers can relate to that moment. But the question comes down to: did you ship something well-built that serves a purpose? If yes, the tooling you used to get there matters a lot less than you think.
Here’s the workflow that actually works:
- Choose your frameworks and specify your stack up front
- Work in small, reviewable chunks – not overnight agent runs
- Use spec-driven development with tests and architecture decisions
- Review all AI-generated output before shipping
Will AI kill open source? The verdict
Micro open source is probably toast. The big building blocks aren’t going anywhere. But the key is to stay in the architect’s seat – choose your frameworks, specify your stack, review the output. Be the architect handing specs to the contractor, and don’t give that role away to the AI.
There’s a lot more in the full conversation including anti-AI vigilante groups shaming people for publishing agent-assisted packages, the open source gift economy, and why none of us really know where this is all heading yet.
