AI coding agents are increasing code and pull-request volume without speeding up ticket completion, according to an analysis of data from 718 companies tracked by Jellyfish. The reported result is a shift in engineering work: less time writing code, more time reviewing it.
Watch Desk analysis
What happened
In a post published on 17 October, the author says Jellyfish’s data shows more lines of code and pull requests per engineer at companies adopting AI coding agents, but no acceleration in closed Jira tickets. The analysis also says the extra code is creating a review bottleneck, lengthening the time needed to merge pull requests.
The post argues that this challenges executive expectations that AI coding tools would quickly reduce hiring needs. The supplied account gives no underlying figures or breakdown of the companies, so the findings should be read as the author’s account of the tracked data, not a universal verdict on every engineering team. Read the analysis.
Why it matters
More code is not automatically more finished work. If generated contributions pile up faster than teams can review and merge them, the pressure moves downstream rather than disappearing. That has practical consequences for how engineering teams assess productivity, staffing and the cost of adopting coding agents.
The distinction is useful for anyone weighing an AI coding tool: output volume may be an easy number to celebrate, while completed work and review time are closer to the actual delivery question. The reported analysis suggests those measures can tell very different stories.
Our read
This is a welcome nudge away from counting lines as if they were shipped value. The data described here raises a sharper question than whether AI can write code: can a team review, integrate and maintain the extra output? Before treating a bigger pile of pull requests as a productivity win, managers should check whether work is reaching users any faster.
What to watch
- Whether the analysis publishes figures on review and merge times.
- How results vary by team size, coding-agent use and type of work.
- Whether later data shows faster ticket completion, or simply a larger review queue.
Discussion spark: If coding agents increase output but not completed tickets, should teams measure their success by code produced, review time or work shipped?
Sources and evidence
- Thank You for Mandating AI. We Still Work Here. (17 October 2026, 11:41 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.