Percona CEO Peter Farkas says the database industry was caught unprepared by AI’s iterative workloads, in which agents may generate and assess around 150 attempts before selecting a result. Existing databases can support these applications, but he believes none is currently an ideal fit and that a purpose-built system may eventually be needed.

Farkas argues that the industry first needs to understand how agentic workloads will develop before designing a clean-slate database. Organisations are still deciding how to deploy AI, whether to run models themselves or use cloud services, and whether smaller models could provide better efficiency than large language models; meanwhile, established databases such as MySQL and PostgreSQL benefit from decades of maturity.