Companies are rapidly embracing AI coding tools, and the payoff is showing up in how quickly software gets shipped.
That’s the central finding of a sweeping new benchmark study from engineering intelligence platform Jellyfish, which analyzed data from more than 700 companies, 200,000 engineers, and 20 million pull requests.
Adoption is already mainstream. Across companies in the dataset, median AI tool adoption stands at 63%, and 64% of companies are now generating a majority of their code with AI assistance. Weekly usage rates have steadily climbed over the past year, with a growing share of engineers using AI coding tools multiple days per week, according to Jellyfish’s data.
The biggest impact is on output. Companies with the highest levels of AI adoption — defined as having 75% to 100% of engineers using AI coding tools three or more days per week — merged an average of 2.2 pull requests per engineer per week. That’s nearly double the 1.12 weekly pull requests at low-adoption companies (A pull request, or PR, is a proposal to add or change code in a shared software project so others can review and approve it before it goes live).
“Lots of folks are accepting that AI helps you code faster and are starting to think about all of the follow-on problems,” Nicholas Arcolano, head of research at Jellyfish, told me in an interview. “Are we actually getting a good ROI from those gains? How’s it affecting quality?”
“Claude Christmas”
He said he stopped writing code himself in the fall of 2025, turning that job over to AI tools. Around that time, big model improvements made AI coding tools a lot better, and usage really took off. Arcolano calls that moment “Claude Christmas,” when many software engineers discovered the real power of Anthropic’s Claude Code service and started playing around with it over the holidays.
“Last Fall was around the time I gave up fully writing my own code,” Arcolano told me. “So I haven’t written code myself, or even really looked at it, since, say, October.”
Other popular AI coding tools used by engineers include OpenAI’s Codex, Cursor, and GitHub Copilot.
Code quality
Importantly, code quality doesn’t appear to be collapsing under the weight of faster output, according to Jellyfish’s data. Revert rates — code that must be rolled back after deployment — tick up only modestly as AI adoption among software engineers rises, moving from 0.61% at low-adoption companies to 0.65% at the highest tier.
“We haven’t seen a lot of quality impacts,” Arcolano said. “That’ll be the thing to watch as we move forward. The quality concerns and trying to hold the line there with mounting pressure to do things faster, and the ability to validate the AI code being a bottleneck.”
Autonomous coding
Meanwhile, more advanced uses of AI are emerging. Autonomous agent activity (pull requests opened or committed by AI agents) remains a small share of overall work but is climbing rapidly, particularly among top adopters.
Arcolano was most struck by the differences between organizations that are fully embracing AI agents for coding and others that are doing very little or nothing in this new field.
“The separation is accelerating,” he told me. “So the people at the bottom are not moving, the people in the middle are moving gradually. The people at the top are on a rocket ship, and they’re running away with it.”
“This is the story of AI adoption,” he added. “This is why there’s such excitement around autonomous agents.”
Sign up for BI’s Tech Memo newsletter here. Reach out to me via email at [email protected].
