Perplexity Engineers Build Custom Database Using AI Agents

Accelerating Development Through Automation

Perplexity recently engineered a proprietary database system called CobbleDB to replace its reliance on Amazon DynamoDB. Developed by two engineers over a two-month period, the system utilizes approximately 40,000 lines of Rust code. The project relied heavily on hundreds of AI coding agents to assist with the complex development process.

The company sought an alternative to DynamoDB due to rising operational costs and insufficient control over read performance. By moving to an in-house solution, the team aimed to optimize data retrieval speeds and reduce overhead. This internal infrastructure now manages a significant portion of the platform's key-value storage requirements.

The rapid creation of CobbleDB highlights a shift in how software teams approach infrastructure projects. By leveraging AI agents throughout the coding phase, the two-person team managed to complete a production-ready database in record time. This method allowed the engineers to focus on high-level architecture while the agents handled repetitive implementation tasks.

Can AI Agents Replace Traditional Engineering Teams?

The resulting system is a specialized key-value store designed specifically for the company's unique performance needs. It offers a level of customization that standard cloud-managed databases often lack. This transition marks a strategic move toward greater technical independence for the search startup.

The success of this project suggests that AI-assisted development is becoming a viable path for building complex backend systems. While the human engineers provided the vision and final oversight, the agents significantly increased the output volume. This collaborative approach allowed a small team to achieve results that might typically require a much larger department.

Frequently Asked Questions

As Perplexity continues to scale, CobbleDB will likely serve as a blueprint for future infrastructure improvements. The ability to build and maintain custom solutions could provide the company with a distinct performance edge. Future updates will determine if this proprietary database can handle the increasing demands of the platform's growing user base.

Why did the company move away from DynamoDB? The company found DynamoDB too expensive and lacked the necessary control over read performance. Building an internal system allowed for better cost management and optimization.

How did AI agents contribute to the project? Hundreds of coding agents assisted the two engineers by writing and testing portions of the 40,000-line Rust codebase. This automation significantly accelerated the development timeline to just two months.