Upcoming Paper Jam: AI Writes Faster Than Humans Can Review

What happens when a company simply orders its engineers to ship twice as much? We'll find out in the upcoming Paper Jam #20. In the paper we'll be discussing, a mid-sized, AI-forward company committed to doubling merged pull requests per engineer from mid-2025 onward. The study covers 802 developers and nearly 200,000 pull requests and shows that per-capita throughput really did reach 2.09x the old baseline.

But, as output doubled, the work didn't vanish — it moved. Per-reviewer load roughly doubled too, and automated review quietly overtook human review. Merge and revert rates held steady, which you can read as reassuring or as unsettling, depending on how much you trust one machine to catch what another machine wrote.

If the framing feels familiar, it should: three of the authors — Hao He, Shyam Agarwal, and Bogdan Vasilescu — also wrote Speed at the Cost of Quality, the Cursor study we discussed back in February for Paper Jam #14. They're picking up right where that conversation left off, and letting us ask where the bottleneck lands once writing the code stops being the hard part.

We'll be meeting on Friday, August 28 at 1pm UTC. This is the paper we'll be discussing:

Hao He, Shyam Agarwal, Yegor Denisov-Blanch, Pavel Azaletskiy, Sanmi Koyejo, Bogdan Vasilescu
AI Writes Faster Than Humans Can Review: A Longitudinal Study of an Enterprise 2x Mandate (2026) [link]

Paper Jams is a monthly one-hour video call where a small group discusses an academic paper on the human factors of making software. We follow the Chatham House Rule, and the conversation goes wherever the group takes it.

Interested? Subscribe to Paper Jams to get invited to the call.