AI Writes Faster Than Humans Can Review: A Longitudinal Study of an Enterprise 2x Mandate
This longitudinal study of a mid-sized enterprise's "2x" mandate reveals that while AI coding tools catalyzed a sustained doubling of per-capita developer throughput by April 2026, the gains were driven by adoption and accumulated usage rather than the mandate itself, resulting in a restructuring of code review workflows where automated processes overtook human review without compromising merge stability.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a software company as a busy kitchen. For years, the chefs (developers) have been cooking meals (code) by hand. Then, in mid-2025, the head chef (the CTO) made a bold promise: "We are going to double the number of meals we serve every month using our new AI kitchen assistants."
This paper is a long-term report card on how that promise actually played out over two years. Here is the story, broken down into simple parts.
1. The Big Result: They Did Double the Output, But Not Instantly
The company hit their goal. By April 2026, the average chef was indeed producing about twice as many meals as they were before the AI mandate.
However, it wasn't like flipping a switch.
- The "Magic Wand" Myth: The paper says the gain didn't happen overnight. It was more like learning to ride a bike with training wheels. At first, the chefs were just getting used to the new tools.
- The "Practice Makes Perfect" Reality: The more the chefs used the AI, the faster they got. The biggest boost came from accumulated use. It took about nine months of heavy use for a single chef to reach that "double the output" mark. The mandate acted like a coach shouting, "Get on the bike!" but the speed came from the chefs actually pedaling and getting better at it over time.
2. Who Got Faster? Everyone, But Especially the New Stuff
- The Whole Team: The speed boost wasn't just for the junior chefs or the masters. Everyone from the newest hires to the top bosses got faster. Interestingly, the managers (who rarely cooked before) saw the biggest percentage jump because they finally had an easy way to jump back into cooking without getting overwhelmed.
- Old vs. New Recipes: The AI worked wonders on new recipes (new code). But when chefs tried to use AI on the old, dusty recipe books (legacy code), it didn't help much. The AI was great at building new wings of the house, but it struggled to renovate the crumbling foundation.
- The "Model" Mystery: The company upgraded their AI brain three times during the study (like upgrading from a smartphone to a super-computer). Surprisingly, the paper couldn't tell which specific upgrade made the chefs faster. It seems the chefs just got better at using the tools, regardless of which specific version they were holding.
3. The Bottleneck: The Kitchen Got Crowded
Here is the twist: The chefs started cooking twice as fast, but the taste-testers (code reviewers) didn't get faster.
- The Traffic Jam: Because the chefs were churning out meals so quickly, the line of food waiting to be tasted got huge. The reviewers were suddenly overwhelmed, with their workload doubling.
- The Robot Taste-Testers: To fix this, the kitchen started using robot taste-testers (automated AI reviews). Soon, the robots were doing more of the checking than the human chefs.
- The "Silent Approval": Humans started doing less "tasting" and more "nodding." They stopped writing long notes about what was wrong and just clicked "Approve" to keep the line moving.
- Did the Food Get Bad? Surprisingly, no. The number of meals sent out (merged) stayed high, and the number of meals sent back to the kitchen (reverted) didn't go up. The food seemed fine at the moment it left the kitchen. However, the paper warns that we don't know if the food will spoil later (long-term quality issues), because the robots might miss subtle problems that humans would catch.
4. The Main Lesson: It's a Process Change, Not Just a Tool
The paper concludes that the company didn't just "install a faster tool." They accidentally redesigned their entire kitchen workflow.
- The Shift: The work didn't disappear; it just moved. The chefs spent less time writing code and more time directing the AI. The reviewers spent less time critiquing and more time managing the robots.
- The Warning: If you just tell people to "go faster" without giving them time to learn the new tools, you create a "productivity pressure paradox" where everyone gets stressed and nothing actually improves. The real magic happened because the chefs had time to learn, practice, and accumulate experience with the AI.
In a nutshell: The company successfully doubled its output, but it took nine months of practice to get there. The result wasn't just "faster coding"; it was a fundamental shift where robots started doing the heavy lifting of checking the work, and humans became more like managers of the robots. The food tasted fine for now, but the kitchen looks very different than it did before.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.