How monday.com Used AI to Shrink a 8-Year Monolith Split into 6 Months

At monday.com, we faced a monumental challenge that seemed almost insurmountable at first-breaking apart our massive JavaScript client monolith. This task was originally estimated to take 8 person-years of manual effort. However, with the development of Morphex, our AI-powered migration system, we managed to complete it in just 6 months. In this article, I'll dive into how we leveraged AI for one of our most complex engineering projects and share some key learnings along the way.

The Moonshot

The journey to Morphex began during an "AI Month" initiative at monday.com. For that month, almost every engineer in the company worked on either building internal AI tools or integrating AI capabilities into our product. One of the task forces we formed aimed to tackle a particularly ambitious goal: splitting our giant client-side monolith and rebuilding it with a modern stack.

A monolith is a single codebase that often holds an overwhelming amount of code, dependencies, and internal complexity. Our client-side application was no exception-it was vast, distributed, and extensible, allowing hundreds of internal developers and thousands of community developers to enhance it daily. It featured a centralized state system based on Redux, with thousands of files containing countless actions, selectors, reducers, constants, services, and utils.

Our goal was ambitious: untangle and rewrite the monolith from JavaScript to TypeScript while migrating to a Zustand-based state management system in just 6 person-months.

Becoming Managers of a Team of Agents

As we embarked on this journey, several questions and concerns arose:

• How do we map out the codebase and its dependencies into work items?
• Can our bots work in parallel without conflicts?
• How do we maintain visibility over our progress?

We decided to use monday.com's own platform to organize and manage the process. Here’s how we tackled each challenge:

Mapping Out the Codebase

• Dependency Analysis: We used static code analysis tools to map out the dependencies within the monolith. This helped us identify which parts of the codebase could be split and migrated independently.
• Work Item Creation: Based on the dependency analysis, we created detailed work items for each component that needed to be refactored. These were then assigned to our AI agents.

Ensuring Parallel Work

• Conflict Resolution: We implemented a version control system that allowed multiple bots to work in parallel. The system automatically resolved conflicts by merging changes and flagging any issues that required human intervention.
• Task Prioritization: We prioritized tasks based on the complexity and dependencies of each component. This ensured that critical parts of the codebase were migrated first, reducing the risk of downstream issues.

Maintaining Visibility

• Progress Tracking: We used monday.com's project management features to track the progress of each work item in real-time. This allowed us to monitor the overall status of the migration and make data-driven decisions.
• Regular Check-ins: We held regular check-in meetings to review the progress, address any issues, and adjust our approach as needed.

Key Learnings

• AI is a Tool, Not a Silver Bullet: While AI significantly accelerated the process, it still required human oversight. Our engineers played a crucial role in setting up the initial work items, resolving conflicts, and ensuring the quality of the migrated code.
• Modularization is Key: Breaking down the monolith into smaller, manageable components was essential for successful migration. This approach not only made the task more feasible but also improved the overall architecture of our application.
• Continuous Learning: The process of building Morphex taught us a lot about AI and its applications in software engineering. We continuously refined our tools and practices based on what we learned.

Conclusion

The development of Morphex was a significant milestone for monday.com, demonstrating how AI can be effectively used to tackle complex engineering challenges. By leveraging AI, we were able to reduce an 8-year task to just 6 months, while also modernizing our codebase and improving our development processes.