Bounties Don’t Save Open Source
AI is getting very good at writing code. Claude Opus 4.5 scored higher than any human on Anthropic’s internal performance engineering exam. Anyone can generate a working PR now. Writing code is no longer the bottleneck.
Review is.
The obvious solution: pay reviewers. Bounties.
But this doesn’t actually work.
Why Bounties Fail
Most qualified reviewers have full-time jobs. Consider the two cases:
Case 1: Passionate Maintainer
They care about the project. They’d review PRs for free if they had time. The constraint isn’t motivation—it’s hours in the day.
Bounties don’t help because:
- They can’t quit their job for sporadic, unpredictable bounty income
- No single repo generates enough PR volume for full-time work
- Even $500/review doesn’t create more hours—they still have the same 24
Money can’t buy time from people who don’t have it to sell.
Case 2: Unpassionate Maintainer
They’ve moved on. The project is abandoned or low-priority. Money might motivate them short-term—but this is unstable:
- Extrinsic motivation produces lower quality than intrinsic
- They’re doing minimum viable review to collect payment
- The moment bounties dry up, so does their attention
The real solution here isn’t bounties. It’s transferring maintainership to someone who actually cares. Paying someone to pretend to care is worse than finding someone who genuinely does.
The Better Approach
Instead of paying for reviews, reduce the need for extensive review.
How? Make the AI’s work easier to verify:
- Runnable examples in CI. If the PR includes tests that demonstrate correctness, review becomes verification.
- Effective integration tests. Tests that actually convince a reviewer the change works.
- Clear, minimal diffs. Less to review means faster review.
The goal isn’t compensating reviewers for cognitive load. It’s eliminating the cognitive load in the first place.
Conclusion
In both cases, bounties fail:
- Passionate maintainers are time-constrained, not money-constrained
- Unpassionate maintainers need replacement, not payment
The bottleneck isn’t money. It’s either time (which can’t be bought) or ownership (which should be transferred).
Focus on making PRs self-evidently correct. No bounties needed.