# AI doubled merged code. Review is where it piles up, LinearB's data says

> Across 2.7 million pull requests, AI-assisted changes were 2.6 times bigger and waited five times longer for review. LinearB now wants to feed that context to agents.

- URL: https://devisdead.ai/news/ai-doubled-merged-code-review-is-where-it-piles-up-linearb-s-data-says
- Published: 2026-11-02T19:51:09Z
- Author: DevIsDead Newsroom
- Section: Dev
- Tags: linearb, code-review, mcp, agents
- Originally reported by SiliconANGLE (https://siliconangle.com/2026/11/02/with-agentic-studio-linearb-provides-the-critical-context-needed-to-accelerate-software-delivery/).

Writing code got cheap. Shipping it didn't. That's the case LinearB made on Monday while launching Agentic Studio, and the numbers behind it are more interesting than the product, as [SiliconANGLE](https://siliconangle.com/2026/11/02/with-agentic-studio-linearb-provides-the-critical-context-needed-to-accelerate-software-delivery/)'s Mike Wheatley reported.

## The data

LinearB's Software Engineering Benchmarks report covers more than 2.7 million pull requests from 250 organizations. Teams that rely on AI coding tools nearly doubled their merged changes year over year. But AI-assisted PRs were about 2.6 times larger and waited more than five times longer for review. And at organizations LinearB rates as elite, changes proposed by agents were merged 79% of the time, against 92% for code written by people.

So the bottleneck moved downstream, to the humans who have to read it all.

## The product

Agentic Studio pulls data from Git repositories and project trackers into one context layer that engineers can question in plain language, asking why releases slowed down instead of reading charts. It also hands that context to coding agents over the Model Context Protocol: rework rates, past incidents and code ownership, so an agent knows a file's history before touching it.

The first agent skill, in early access, decides effort. It weighs a PR's complexity, rework history and risk, then sets how much reasoning and testing an agent spends on it: cheap, quick passes for simple changes, deeper work for sensitive ones. Skills for code review, merge governance and token costs are planned.

The showcase customer is Kraken, which used the MCP integration and found that time-zone gaps, not PR size, were slowing its reviews. After reorganizing, its review times fell by more than half in four months.

## Benchmarks vs. vibes

These are a vendor's numbers, used to sell a fix for the problem they describe, so read them as directional. They do match what many teams report: more code, bigger diffs, slower reviews. The 13-point merge gap is the stat to watch. Every agent PR that sits in review and then gets rejected is human time spent on code that never ships. And note that Kraken's win came from changing how people worked, not from adding more AI.

**[Scoreboard](/scoreboard) update:** Humans +1. Code written by people still gets merged more often than the agents’ (92% vs 79%).

Stay undead.