How Cresta brings consistent review to AI-generated code with CodeRabbit

Company snapshot

Cresta develops enterprise AI for customer experience. Its platform combines AI agents, real-time assistance for human agents, and conversation intelligence to help businesses support customers, coach employees, and improve service operations.

Founded in 2017 out of Stanford’s AI Lab, Cresta serves enterprises including United Airlines, Marriott, and Cox Communications. The company has more than 600 employees across more than 10 global team hubs.

Its engineers build both the underlying platform and the implementations that connect Cresta’s AI agents to customers’ businesses. Reliability and compatibility are essential as those products evolve.

Development environment

GitHub, with AI-assisted development and core services written in Python and Go

Why CodeRabbit

Useful findings, independent review, repository-specific configuration, code walkthroughs, and sequence diagrams.

The result

More consistent reviews that catch bugs, flag potential breaking changes, and give senior engineers code with initial findings already addressed. Cresta increased its contracted CodeRabbit seats by 40%, from 150 to 210 Enterprise seats, within four months.

Applying shared standards as code volume grows

For Cresta’s globally distributed engineering organization, consistent review depended heavily on the knowledge of individual engineers. Experienced reviewers understood team conventions and the requirements of particular services. As AI coding tools increased the volume of changes, Cresta wanted to make that guidance a repeatable part of review.

Xing helped assess how automated review could support that work. The company wanted higher code quality, more bugs caught before merge, and an independent reviewer for AI-generated changes. It also wanted feedback early enough for developers to act before CI or human review, with configurable checks that kept comments useful and relevant.

Cresta compared CodeRabbit with other review approaches, examining the quality of findings, cost, and the experience of using each product. CodeRabbit’s walkthroughs and sequence diagrams helped explain changes, while repository-level configuration let teams specify how their code should be reviewed. Together, those features helped it earn support among the engineers participating in the selection.

Xing also valued CodeRabbit’s handling of Python and Go, languages central to Cresta’s work. Choosing a specialized reviewer gave Cresta access to capabilities that would otherwise require its own engineers to develop and maintain.

Establishing the results in real pull requests

Cresta began its CodeRabbit evaluation with two repositories in January 2026. The rollout included GitHub reviews and local development workflows, supported by organizational configuration, path-specific instructions, and coding guidelines.

With limited baseline data on review time and quality, Cresta examined the bugs and meaningful issues CodeRabbit surfaced before merge, whether developers accepted its suggestions, and how experienced reviewers judged the resulting code.

During the initial month of use, CodeRabbit covered 1,871 pull requests, and developers accepted 51% of its suggestions, saving an estimated 2.6 weeks of cumulative reviewer time.

Cresta’s leaders saw the benefit in how consistently changes were being examined.

“Applying the best level of review consistently across the company — that's the biggest gain,” said Daniel Hoske, Cresta’s Chief Technology Officer.

Hoske also said some of Cresta’s most stringent reviewers had little left to add after developers addressed CodeRabbit’s findings.

Catching a critical React rendering issue

Xing shared a colleague’s example in which CodeRabbit identified a recursive rendering issue in a React pull request. Entering a particular value could leave the interface unresponsive.

“It’s actually a critical bug for the front end,” Xing said.

Xing said the engineer had overlooked another interaction in the code. CodeRabbit’s finding gave the team an opportunity to address the issue before customers encountered it.

Building Cresta’s requirements into review

On Cresta’s platform teams, CodeRabbit reviews changes before engineers ask colleagues to examine them. Authors can address its findings first, making automated review a regular part of preparing a pull request for human attention.

“I use CodeRabbit in all of my PRs,” Xing said.

His team has also configured the reviewer for specific requirements. One particularly useful application involves a command-line interface the team maintains. As engineers update the CLI, they use CodeRabbit to check backward compatibility and flag potential breaking changes.

Beyond the CLI checks, teams use repository instructions and review conversations to explain their coding conventions and requirements.

Xing observed that engineers have become more inclined to write down best practices and style conventions because they know CodeRabbit can use that information. Guidance that once depended on an individual reviewer remembering a rule can become part of future automated reviews.

During Cresta’s first several months of use, Xing also saw improvements in CodeRabbit’s interface for managing knowledge and repository preferences, giving teams more control over those instructions.

Cresta also uses CodeRabbit’s pull request summaries to help prepare changelogs and release notices, giving customers an overview of changes to its services.

Supporting engineers who build customer AI implementations

Cresta’s forward deployed engineers also use CodeRabbit as they adapt the company’s AI agents to individual customers’ businesses.

These engineers work with customers to understand their operations, then build the tools, hooks, and processing logic required for an implementation. They may support several accounts at once, with individual implementations delivered in as little as one or two weeks.

The team uses AI coding tools extensively, and its code draws on Cresta’s own agent framework. According to Xing, substantial documentation in the repositories helps CodeRabbit understand that code and its intended use.

Automated code review fits alongside Cresta’s evaluation framework, which tests how the resulting AI agents behave. Engineers can examine the implementation code and evaluate the agent’s performance as they prepare a customer deployment.

Expanding CodeRabbit across engineering

Cresta purchased 150 CodeRabbit Enterprise seats on March 30, 2026. It added 20 seats in April and another 40 in July, bringing the contracted total to 210.

Hoske had recognized the benefits early in Cresta’s adoption.

“We do see the value quite significantly from CodeRabbit,” he said.

Xing uses CodeRabbit on every pull request he submits. As AI coding tools take on more implementation work, he sees automated review as a required part of the development process.

“AI code review is absolutely necessary.”