Taskrabbit cut merge time by 25% before adopting coding agents

Taskrabbit has experienced significant growth since its launch over a decade ago. Because of that growth, the popular platform that connects people with skilled local “Taskers” has a large engineering team and is always seeking innovative ways to streamline its processes and make its developers’ lives better and more productive.

Kiran Kanagasekar, a Senior Engineering Manager at Taskrabbit, leads multiple teams, including the Developer Excellence team, which aims to enhance developer productivity across the organization. While many companies implemented AI coding agents to enhance productivity, Kiran zeroed in on code reviews as the critical area needing improvement, as they were an existing bottleneck that was affecting their development cycle time.

He believed that jumping straight to implementing AI coding assistants would only exacerbate the existing review bottleneck that saw consistently slow merge times – and not make the company any more productive. So, Kiran focused on leveraging AI for code reviews. This move led Taskrabbit to CodeRabbit, an AI tool that promised to fix their code review problem.

Challenge: Slow time-to-merge and inconsistent code standards

Before using CodeRabbit, Taskrabbit faced common code review challenges that hurt their merge time and frustrated developers. These issues worsened due to their global engineering team and massive, monolithic codebase.

• Slow review cycle due to global teams

Taskrabbit’s engineering team is spread across the globe, with teams in the US and Poland. This global setup meant that if you created a pull request late in your time zone, you’d be waiting for feedback until the next business day in a different time zone. Kiran put it this way:

If I create a PR after 2 o’clock in the Pacific time zone, that’s it. I don’t receive any reviews from Eastern Time or European people. I have to wait for the whole day, and then they’ll come back and provide a comment.

This built-in delay, resulting from the company’s geographic spread, directly led to longer development cycle times.

• Inconsistent coding standards across a growing codebase

Maintaining consistent coding standards across a decade-old, evolving codebase was a persistent struggle for Taskrabbit. New senior engineers, despite their general industry knowledge, often lacked immediate familiarity with Taskrabbit's unique patterns and anti-patterns. This disparity led to inconsistencies in code reviews and approvals. While Taskrabbit had attempted to address this through linters and automation using the Danger framework, they recognized the need for an innovative, learning-based code review tool that could consistently enforce best practices.

• Potential for delayed identification of vulnerabilities and optimizations

While Taskrabbit didn’t ship many bugs, the manual review process occasionally overlooked vulnerabilities or opportunities for code optimizations. Leveraging CodeRabbit as a first-pass reviewer allowed them to streamline the review process for senior reviewers, who could then focus on more substantive issues.

• AI coding tools would have worsened the merge time bottleneck

Kiran understood that adding AI coding assistants would create more PRs and worsen the review process.

Writing code faster was never the issue; the bottleneck was always code review,

• Kiran Kanagasekar, Senior Engineering Manager

Kiran’s foresight prevented Taskrabbit from simply exacerbating the bottleneck and frustrating the team.