How Cadent uses CodeRabbit to scale AI-assisted development and align engineering teams

Company snapshot

Cadent operates a predictive advertising platform that connects brands, publishers, and consumers across digital media, linear television, connected TV, and other channels. Its platform makes an estimated 200 billion connections each day and employs 22,000 AI-driven models to help advertisers reach audiences and improve campaign outcomes.

Supporting that platform requires Cadent's engineers to deliver code quickly while protecting the quality and stability of systems operating at enormous scale. That responsibility falls partly to Chris Gallagher, Cadent's Senior Director of DevOps and Cloud Engineering.

"My job is to make sure developers can work efficiently and safely to get good code out the door as fast as possible," Gallagher said. "If you're late to market, you're just not going to win that business."

AI accelerated coding moved the bottleneck to review

Cadent's developers use Cursor, Claude, and other AI tools to write code. These tools increased development speed, but they also produced changes that were difficult for human reviewers to process.

In one project involving internal AI capabilities, developers regularly submitted changes spanning 22,000 lines and 75 files. In another instance, a developer generated 70,000 lines of changes in a week to support three SQL statements.

"Nobody can check that realistically," Gallagher said. "Once teams realized they were spending two or three days looking at AI-generated code, they knew they couldn't keep up with that pace."

Cadent examined its delivery pipeline and found that pull requests had become the largest source of delay. Gallagher estimated that the PR stage accounted for roughly 60% of the time between code check-in and delivery. Large reviews sat unclaimed because senior developers knew how much time they would require.

That delay affected more than individual pull requests. Cadent's strongest developers were spending their time working through repetitive review tasks instead of writing code, designing new systems, and building the next set of product capabilities.

"If developers are stuck in code review cycles for three, four, or five days, nothing is getting out the door," Gallagher said. "Then you start getting a backlog, and it gets worse and worse."

Choosing an independent reviewer for AI-generated code

Cadent evaluated several AI review tools, including review capabilities bundled with AI coding platforms. The company chose CodeRabbit because it provided more complete review capabilities and an independent perspective on code generated by other AI systems.

Gallagher compared the decision to the question, "Who watches the Watchmen?" If the same system writes and reviews the code, it may reinforce its own assumptions or overlook the same weaknesses twice.

Cadent wanted separation between code generation and code review. Its proof-of-concept testing found that CodeRabbit delivered more useful feedback than the alternatives.

"We did the CodeRabbit demo and a few tests and POCs, and we got better feedback," Gallagher said. "That was the instant thing for us. It was easy."

Cadent did not set out to remove human judgment from review. Instead, the company wanted CodeRabbit to handle the repeatable checks that consume reviewer attention, including formatting, linting, straightforward mistakes, and common quality problems. Human reviewers could then focus on business logic, system behavior, and the decisions that require knowledge of Cadent's products.

"We wanted it to do the 90% of automated things that every developer does during code review," Gallagher said. "If it can take care of the basics and leave us with the business logic, why would we not do that?"

Catching defects before they become operational problems

CodeRabbit began producing concrete results across Cadent's engineering organization.

Gallagher's DevOps and cloud engineering team estimates that CodeRabbit routinely caught Terraform mistakes. These included formatting errors that would have broken builds and infrastructure-as-code findings that could have caused problems later in the delivery process.

Cadent also connected its global CodeRabbit configuration with findings from the Wiz security platform. This allowed CodeRabbit to identify several infrastructure issues before deployment.

What's more, CodeRabbit identified a code path that led nowhere and would eventually have produced a difficult-to-diagnose bug. Gallagher said the problem sat deep enough in the code that the team might not have discovered it for another month or two.

Cadent's data engineering organization has caught logic errors before release. Developers working on a complex, 20-year-old application use CodeRabbit to review changes and improve nearby legacy code while they work. Because the feedback identifies the relevant lines and suggests a remedy, developers find it easier to act on than traditional static-analysis reports.

Cadent's internal data also shows fewer bugs reaching production from teams that use CodeRabbit more extensively.

"We can track the applications we release and the number of bugs that go out the door," Gallagher said. "We're seeing a nice downswing from the teams using CodeRabbit more than the teams that aren't."

More than half of CodeRabbit's feedback is accepted

A recent Cadent usage snapshot shows CodeRabbit operating across a broad portion of its engineering environment.

For the last three months, there was an overall CodeRabbit acceptance rate of 57.8%.

The results were especially strong for higher-severity feedback. Critical findings achieved a 70.8% acceptance rate. Altogether, critical and major issues accounted for approximately 69% of all accepted comments.

CodeRabbit saved nearly two weeks of reviewer time during the period. Gallagher sees that time returning to the senior engineers who previously carried much of Cadent's review burden.

"They're not spending as much time in code review hell," he said. "They're able to go back and write code or design the next level of code and the next features and functionality."

Turning review feedback into better AI-generated code

CodeRabbit's feedback has also changed how Cadent's developers use AI coding tools.

When CodeRabbit identifies a recurring problem in Cursor-generated code, developers can update their Cursor rules, skills, or other instructions to prevent the coding agent from making the same mistake again.

That creates a feedback loop between generation and review:

Developers produce code with an AI assistant. CodeRabbit reviews the changes independently. Teams then apply the lessons from that review to improve future AI-generated code.

"It's a little extra work up front, but you get it done," Gallagher said. "Then it eliminates a lot and speeds up that portion of the process."

This separation gives Cadent the productivity benefits of AI-assisted development while maintaining an independent quality checkpoint.

Bringing acquired engineering teams into a shared development culture

Acquisitions have played an important role in Cadent's growth. During Gallagher's nearly eight years with the company, Cadent has acquired roughly one company per year. Each acquisition brings valuable technology and engineering talent, along with its own repositories, tools, coding practices, and development culture.

Those differences can create silos. Teams may solve similar problems in different ways, follow different standards, or lack visibility into lessons already learned elsewhere in the company.

CodeRabbit is helping Cadent transfer engineering knowledge across those boundaries.

Cadent has created 531 organizational learnings in CodeRabbit. During the last 90 days, those learnings were applied approximately 39,500 times and reached 64.2% of reviewed PRs. Path-based instructions were applied another 5,120 times across 40.3% of PRs.

These capabilities allow a lesson discovered in one part of Cadent to inform reviews elsewhere. They also help acquired teams understand existing standards without requiring senior engineers to explain the same practices during every review.

Gallagher saw the difference after Cadent completed its first acquisition following the CodeRabbit deployment.

"It's been interesting to see how much smoother that integration has been," he said. "They started getting useful feedback, and then they went to people writing similar things and asked, 'Could you show me why you were doing it that way?' Before, everybody stayed siloed. Now we're seeing more speed toward, 'We all work together.'"

Creating consistency without removing team autonomy

The same effect is emerging across Cadent's existing development groups.

Cadent has approximately a dozen Python teams that historically used different linting practices. Gallagher's team introduced a global CodeRabbit rule as a starting point. The rule prompted the Python team leads to discuss which differences were genuinely necessary and where they could adopt shared standards.

That conversation produced benefits beyond cleaner code. Greater consistency makes it easier for developers to move between teams, understand one another's work, and help wherever additional engineering capacity is needed.

Teams control their own requirements. They can create repository-specific CodeRabbit configuration files when a project needs different treatment. In practice, many developers have asked Gallagher to add useful rules to the global configuration so other teams can benefit from them.

"We're seeing more collaboration," Gallagher said. "That was a shocking finding, to be honest. I didn't expect that, so I'm very happy with it."

Cadent is now preparing team leads to take greater ownership of their CodeRabbit configurations. The company is also expanding enablement around AI-assisted coding and review, building on a global foundation while giving individual teams the context and control they need.

Expanding from 75 to 125 users

CodeRabbit adoption has increasingly spread through developer demand. Cadent initially had roughly 40 to 50 active users within a 75-seat deployment. As individual developers gained experience with the product, their teammates began asking for access.

"A few teams had one person use it and have a good experience," Gallagher said. "Then they were saying, 'All of us. We all want it.'"

Cadent is also preparing to expand its deployment from 75 to 125 users. The company plans to continue consolidating repositories in GitHub while supporting the code that still resides in Bitbucket, Azure DevOps, and GitLab.

The clearest sign of CodeRabbit's value emerged during renewal discussions. When developers learned that an administrative delay might briefly interrupt service, they worried about returning to slower reviews while AI tools continued generating larger volumes of code.

"They said, 'This is going to slow us down significantly,'" Gallagher recalled.

For Cadent, CodeRabbit has become part of the infrastructure required to scale modern software development. It helps developers move faster, gives senior engineers more time for higher-value work, catches defects before they create operational problems, and spreads engineering knowledge across teams that arrived through different paths.

"Every time we find something that improves it, we get a benefit," Gallagher said. "Otherwise, we wouldn't be renewing and increasing our number of seats."