A few years ago, if I send a complex PR that compiles and passes tests, that implies a certain amount of time and cognitive investment on my part. It seems likely that I wouldn't invest that if I didn't also understand the codebase, the feature or bug I'm working on, etc.
Now, that understanding is roughly as expensive as before, but AI has vastly reduced the cost of generating the code that compiles and passes tests.
Probably-well-intentioned community members are happy to contribute the cheap thing( Claude Code tokens) but, because it's so cheap, it's not a good indicator they've contributed the expensive thing (human understanding).
"Writing Code vs. Shipping Code: Productivity Effects Across Generations of AI Coding Tools"
As the FT summarizes,
> They found an explosive impact at the top of this funnel — coders created or edited almost 300 per cent more files — but that boost was halved to 150 per cent by the time they got to the number of discrete pieces of work submitted for review, and that in turn shrunk fivefold to a roughly 30 per cent uplift in the number of full software releases.
https://www.ft.com/content/8e9ae7a4-7209-4e2c-aa36-f3af77d6c...
So as I wrote, AI vastly improves labor productivity on _coding_, but not nearly as much on code _review_ or other parts of the release pipeline.
And, unfortunately, for many open source projects, it's easy for volunteers to send code for review, but hard for them to volunteer reviewing PRs, managing releases, etc.
Yes, this is the takeaway for me. A PR can no longer be a reasonable proof of work.
I see this position a bit: the notion that AI-generated code has no value. I think it's easy to generate zero-value code, but I don't agree that all AI-generated code is zero-value. I've been working on my side projects in OpenCode, and I spend quite a bit of time prompting, setting up the right files, descriptions of the product I'm trying to build, and the roadmap for it. I have a tight validation loop that lets me run through a bunch of automated checks after each change, and then I do a bunch of manual testing through edge cases that the generated feature might screw up, and then I iterate. It's a different kind of work, but I can make progress more quickly than I could coding by hand. Validation loops are the main critical component.
My experience doing this over the past months is that using AI to code is a skill, and I learn new techniques and get better at it as I try stuff. But that also suggests that, when done well, it can produce something of value.
All of this is to say: while I take issue with your first sentence, I completely agree with your second sentence. What we've lost is the ability to distinguish easily between something well-thought out and something generated thoughtlessly. Focusing on cheap validation would help here immensely, as well.
I see all projects moving this direction. Makes more sense to hash out a plan together.
It’s the same about published journal article. A lot of them are a few pages. That is mostly one hour of typing. But everyone knows that typing it is not the work.
Deep research in the codebase, deciding on the flavor of code to write that matches the project, deciding how you'll model the feature with types, how to architect it so that it's testable, writing the tests, foreseeing cases beyond the obvious path, etc.
What changed is that it can be automated. Or, just grant a world where AI is perfect at implementation.
Now our time/energy/attention is freed up to concentrate the work around planning what to build. And the interesting part is that it becomes the input into the AI implementor.
This is a good thing since we tended to skip the planning stage since it's hard in its own way. Or we start building something and then try to synthesize a high level direction from it, yet now since refactoring is so expensive, we're committed to a solution.
Yes, that is exactly what this announcement is about. That it was too much work for them to tell those two apart.
That is exactly the issue. Projects that are end-user applications - as opposed to libraries or development tools - probably see far more slop than actual work like you've described. The yields are too low for it to make any sense to try to figure out which is which, their time is better spent doing the work.
So any project that doesn’t accept AI PRs is really missing out on significant investment
Would you pay 2000$ for those tokens? If not, the number is meaningless.
I'm 100% on the side of maintainers here, but this is BS. If you could "just prompt Claude yourself" the AI productivity boosts would be in hundreds if not thousands of percent, which is demonstrably and self-evidently not the case (at least as of June 2026).