跪拜 Guibai
← All articles
Open Source

ESLint Shuts Down Its Contributor Pool: AI-Generated PRs Made Human Effort Impossible to Verify

By parade岁月 ·
Read original on juejin.cn ↗ Google Translate ↗ Alt translation

Any open-source project that rewards contributions by count or volume will hit the same wall: AI drives the marginal cost of a PR toward zero, so verification becomes a sink of maintainer attention and community trust. ESLint's decision is an early template for how maintainers may respond when they can no longer tell human work from generated output.

Summary

ESLint's contributor pool is dead. Founder Nicholas C. Zakas announced the immediate end of the program on November 13, ending a five-year run that distributed $51,180 to external contributors. The program paid a flat $1,000 a month at first, then 10% of monthly revenue, and it worked: payouts climbed from $7,900 in 2022 to $24,680 in 2025, and some contributors who started with a paid PR eventually joined the team.

Then AI arrived. ESLint, like every open-source project, started receiving a rising tide of AI-generated issues and PRs. The team's first response was a rule: AI-generated contributions were not eligible for rewards. But enforcing that rule meant figuring out which PRs were AI-written. Self-disclosure was requested and ignored. Manual checks got harder as models got better. Community members turned into 'AI vigilantes', commenting under PRs and DMing Zakas to flag accounts they believed were bots. The team finally decided it wasn't worth the time and pulled the plug.

The deeper casualty is the 'pay for human effort' model. The pool existed on the assumption that behind every PR was a person who spent real time and energy; the money was thanks for that investment. AI broke the link between output and effort. When you can't verify effort, you either pay out for nothing or burn maintainer energy doubting contributors. ESLint chose a third path: refund the program. Zakas also noted the suspicion itself corroded the community's enjoyment. Software still gets maintained; v10.15.0 shipped the same day. What's gone is a reward system designed for a pre-AI era.

Takeaways
— ESLint paid external contributors $51,180 over five years.
— The contributor pool started with $1,000 per month and later switched to 10% of monthly revenue.
— 2025 payouts reached $24,680, more than five times the $4,500 paid in 2023.
— AI-generated issues and PRs increased from 2025 onward.
— ESLint required self-disclosure of AI use in PR descriptions, then added manual checks.
— Community members became 'AI vigilantes', flagging suspected AI-written PRs and messaging Zakas directly.
— ESLint ended the program immediately rather than continue verifying PR authorship.
— ESLint v10.15.0 shipped on the same day the termination was announced.
— Zakas said the mechanism was designed for another era and that open source has changed.
Conclusions

The pool's core assumption was that output correlates with human effort; AI severed that correlation, so the reward's meaning disappeared.

Self-disclosure as a rule only works when the incentive to lie is absent; a direct cash reward makes honesty irrational.

The real cost of AI contributions is not the payout but the maintainer time spent policing and the community suspicion that follows.

This will likely push other projects to abandon output-based rewards or to shift toward signals that are harder for AI to fake.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗