Jellyfish March 2026 benchmark 700+ companies

Jellyfish’s March 2026 engineering benchmark adds scale to a claim that has been circulating across software teams for months: AI coding tools are no longer a pilot. The company’s AI Engineering Trends page says the dataset now covers more than 1,000 companies, 200,000 engineers and 37 million pull requests, with results updated through the end of February. Jellyfish says median AI adoption across companies reached 68%, and that the top adopters are seeing about 1.8 times pull-request throughput. The benchmark is presented as a public dataset rather than a one-off survey, combining usage and workflow signals from engineering teams that use Jellyfish’s platform. ### How big is the dataset behind this benchmark? Jellyfish says the March 2026 benchmark covers 37 million pull requests, 200,000 engineers and more than 1,000 companies. The company describes it as the largest study of its kind and says the data is meant to provide a baseline for comparing AI adoption and engineering outcomes across organizations. April 7, 2026 is the date Jellyfish’s head of research, Nicholas Arcolano, used when he wrote that the latest report had been “updated through the end of February” and now covered more than 37 million pull requests. (jellyfish.co) That post tied the public benchmark to a newer release cycle than earlier Jellyfish materials that referenced about 700 companies and 20 million pull requests. ### What does Jellyfish say about AI adoption right now? Jellyfish’s benchmark page says median AI adoption across companies is 68%. The same page says 26% is the share of code generated with AI across companies, while the company tracks adoption through measures including access, weekly active use and frequent use. April 2026 data in Arcolano’s post said the gap between advanced users and the middle of the market is widening. (jellyfish.co) He wrote that the top 10% of companies had moved from 10% to 14.5% adoption of autonomous agents in a short period, while the median company remained around 2%. ### Where does the “AI doubles output” claim come from? Jellyfish’s public benchmark says top AI adopters are posting 1.8 times pull-request throughput. (jellyfish.co) Arcolano wrote separately that companies with strong adoption of AI coding tools are continuing to see “roughly 2x improvements” in pull-request throughput. That framing is narrower than saying every team is doubling output. Arcolano wrote that teams with more modest adoption were seeing gains in the 30% to 60% range, which he said showed value across the adoption curve even before teams move into more autonomous workflows. (jellyfish.co) ### What does the benchmark say about quality and workflow trade-offs? Jellyfish’s April 7 post said reverted code rises as AI use increases, with a 5% to 11% increase in reverted code as usage goes up. (jellyfish.co) Arcolano wrote that the increase was not large enough to offset the throughput gains, but said it remained material enough to watch. The same post said higher output starts to expose other bottlenecks. (jellyfish.co) Arcolano named pull-request reviews, quality assurance and coordination as constraints that become more important as teams increase throughput. ### How does this compare with Jellyfish’s other 2026 research? May 7, 2026 research from Jellyfish based on a separate survey of more than 600 engineering professionals pointed in the same direction, though it used different methods. (jellyfish.co) That report said 64% of respondents believed they were achieving at least a 25% increase in developer velocity and productivity using AI, and quoted CEO Andrew Lau saying aggressive adopters were beginning to “pull away” from more hesitant competitors. The benchmark and the survey are not the same product. The benchmark is based on observed engineering signals and workflow data, while the State of Engineering Management report is based on responses from engineers and managers about productivity, budgets and tool usage. ### What should readers watch next? Jellyfish’s AI Engineering Trends page remains live as a public benchmark, and the company says the dataset is updated over time rather than fixed to a single release. (jellyfish.co) Nicholas Arcolano’s April 7 post points readers to the full dataset, while Jellyfish’s separate 2026 management report offers another checkpoint on adoption, productivity and tool mix across engineering teams. (jellyfish.co)