Perplexity’s AI agents helped build a database. They weren’t allowed to run it.
Perplexity decided it was paying too much for DynamoDB and wasn’t getting the control it wanted over read performance. So The post Perplexity’s AI agents helped build a database. They weren’t allowed to run it. appeared first on The New Stack .
Perplexity AI, facing high costs and limited control over its DynamoDB database, decided to build its own solution called CobbleDB. In just two months, two engineers and hundreds of coding agents developed a 40,000-line Rust key-value store that now handles a portion of Perplexity's production search traffic. Tests showed CobbleDB delivered median batch-read latency of 5.6 milliseconds, significantly faster than DynamoDB's 31.4 milliseconds, while p99 latency improved from 123 milliseconds to 24.2 milliseconds.
The new database is projected to cost at least 20% less than DynamoDB and plans to be open-sourced eventually. Despite the AI agents' help in building CobbleDB, they were not granted access to the production database, as warned by CMU professor Andy Pavlo about the risks of AI agents working with critical production data. The custom database is part of a three-tier storage system, with Pillar storing durable document state, Lorry packaging updates and moving them through S3 to CobbleDB, and Processed page data spread across three replicas per partition for quick access.
While CobbleDB offers faster performance and cost savings, the long-term costs and maintenance of a custom database remain a challenge.
Written by urgent.news from The New Stack's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
