Iterative workloads caught vendors off guard, but requirements for a clean-slate solution remain unclear
INTERVIEW Percona CEO Peter Farkas reckons AI caught the database industry off guard and that agentic workloads may eventually require a new kind of database – once anyone works out what those workloads need.
Farkas spoke to The Register at the company's Percona Live event in Amsterdam, where members of the often-fractious open source database community gathered to discuss best practices, case studies, and the future.
Percona provides software and support across a range of open source database technologies. Over its 20-year history, it has worked with MySQL, MongoDB, PostgreSQL, Valkey, and Redis.
MySQL's first release arrived in 1995. The PostgreSQL project marked its 30th anniversary in July 2026, although its roots in the POSTGRES project stretch back further.
Both databases therefore have decades behind them and represent mature, stable options – just as the conservative world of database administration prefers. The surge in AI services, however, is introducing workloads that these systems were not designed around, and Farkas believes none of the current choices are ideal.
Of the current selection, Farkas said: "It's definitely not the best that we can have. It's the thing that we have right now."
"I think AI definitely caught the industry off guard," he said. Farkas pointed to "iteration-focused workloads" in which an agent might make 150 attempts before a person or another system selects the best result.
"A human would probably not push 150 options just so someone or another AI can pick out the winning one," he said.
For now, organizations must select technology capable of handling these workloads without assuming that today's choice will remain the best fit. Farkas was careful not to favor a particular database.
"Everything needs a database," Farkas said, though coming from the CEO of Percona, that is perhaps unsurprising. "If the new buzzword is blockchain or NFT or cloud or whatever, behind all of those things, there's a database, and AI is no exception.
"AI needs a database, and AI is going to use the databases that we have for the time being, but I do believe that the best solution for AI or agentic workloads does not exist at this point, as there's just not enough time for something to become as mature as the trusted database software."
The rapid pace of change sits uneasily with a database industry accustomed to products maturing over decades. AI development, by contrast, still has the feel of technology being flung at a wall to see what sticks.
Farkas laughed when asked when such a solution might emerge. "I don't have a crystal ball," he said. He believes the AI market must mature before its requirements become clear enough for anyone to build a clean-slate database designed around them.
Organizations are still deciding where AI belongs in their technology stacks, he said, including whether to run models themselves or use cloud services, and whether every use case needs a large language model rather than a smaller model with a lighter footprint.
"It's easy to experiment with something that can just do everything, but that also means that the resources being consumed are outsized compared to the result," Farkas said.
Farkas became CEO of Percona in October 2025, succeeding Ann Schlemmer. He had previously worked at Percona as director of support until 2016. In December 2021, he co-founded FerretDB, an open source service that converts MongoDB wire protocol queries into SQL for PostgreSQL or SQLite.
MongoDB sued FerretDB in May 2025, alleging patent infringement, false advertising, and trademark misuse. FerretDB disputes the allegations.
Asked what lessons he had brought from FerretDB, Farkas replied: "Lots of PTSD from courtrooms."
Farkas also brings experience of working across multiple open source communities – useful as Percona shifts from a company primarily associated with MySQL to a database-agnostic, multi-tech business. Doing so has its own challenges. He noted that different communities work in markedly different ways.
"We had multiple attempts at adding Postgres to our portfolio," he recalled, "but we had some wrong ideas that were based on the MySQL experience on how to engage with that community."
"They have very different preferences. They have very different red flags they are looking for. For example, if you call something 'Percona TDE for Postgres', that is not going to fly: 'What is this nonsense? This is vendor lock-in! This is just one company trying to push their crap!' Whereas in the MySQL space, that's totally normal, and we needed to unlearn these things."
Could Percona eventually extend its open source approach to AI?
"Frankly, AI is a vendor lock-in situation," Farkas replied, pointing to services such as Gemini and ChatGPT.
He argued that there is not yet a mature, enterprise-ready way to run open-weight agents, although Percona is exploring what it might contribute in that area.
"Percona is not anti-AI," he said. "Percona is anti-rebranding Postgres or MongoDB as the AI database." ®