Cockroach Labs Uses AI Agents to Add IBM Db2 Migration Support to MOLT in Two Days

Rapid Db2 Integration with AI

Cockroach Labs successfully added support for IBM Db2, a relational database, to its MOLT migration tool in less than two days. This integration included a new schema converter, Fetch and Verify paths, an ANTLR grammar, and CI Docker images, totaling over ten thousand lines of test fixtures. This rapid development contrasts sharply with the nine months and approximately $160,000 in engineering time required for a similar Oracle support addition in 2024.

AI-Driven Development Process

The entire Db2 integration was completed without human-written code, driven by AI agents. The token cost for this effort was $4,172. The process began with a single GitHub issue, which a planning agent decomposed into fifteen sub-issues with a dependency graph. These sub-issues were further broken down, leading to a total of thirty-two sub-issues, twenty-seven merged pull requests, and fifty-five review cycles. Only two issues required human escalation.

Introducing MOLT Sinai: The Coding Hospital

To manage this AI-driven development, Cockroach Labs built a system named MOLT Sinai, conceptualized as a 'coding hospital.' In this framework, GitHub issues are treated as 'patients,' merging code is 'discharge,' and human oversight is provided by 'Chiefs of Medicine.' This structured approach ensured quality control throughout the automated coding process, with test coverage equivalent to the best-tested dialects like PostgreSQL.

Impact on Development Efficiency

This AI-powered development workflow resulted in a 164-fold increase in speed and a 38-fold reduction in cost compared to traditional methods for similar complex database integrations. The successful implementation of Db2 support demonstrates a potential shift in how complex software features can be developed, leveraging AI agents for significant gains in efficiency and cost-effectiveness.