No Database Built for AI Agents Yet, Says Percona Boss
Peter Farkas, CEO of Percona, thinks the database industry got blindsided by AI. Speaking at Percona Live in Amsterdam, he made no bones about it: the right database for agentic workloads doesn't exist yet, and nobody's building it because nobody's quite sure what it needs to do.
Percona has been around for two decades, working across MySQL, PostgreSQL, MongoDB, Valkey, and Redis. MySQL turns 30 this year. PostgreSQL is already past it. These are mature, battle-tested systems that database administrators trust precisely because they've been around long enough to have their rough edges worn smooth.
Then along comes AI and starts throwing entirely different kinds of workloads at them.
Farkas highlighted what he calls "iteration-focused workloads" as a particular mismatch. An AI agent might generate 150 candidate outputs before another system, or a human, picks the winner. That's not how traditional database usage looks. "A human would probably not push 150 options just so someone or another AI can pick out the winning one," he said. No, they wouldn't. But AI agents do, and existing databases are being pressed into service anyway.
His assessment of the current options was blunt: "It's definitely not the best that we can have. It's the thing that we have right now."
Farkas is careful not to talk down any particular technology. Picking favourites when you run a multi-database support and software company would be a bad look. But his broader point stands: whatever database ends up being genuinely suited to agentic workloads probably hasn't been built yet, and it takes decades for database software to reach the kind of maturity that enterprise users actually trust.
When asked when that purpose-built solution might arrive, he laughed. "I don't have a crystal ball." His view is that the AI market itself needs to settle down first. Organisations are still working out whether to run their own models or outsource to cloud providers, and whether every use case actually warrants a large language model or whether something smaller and cheaper would do the job fine.
"It's easy to experiment with something that can just do everything," he said, "but that also means that the resources being consumed are outsized compared to the result."
Farkas took over as CEO in October 2025, returning to Percona after previously running its support operation until 2016. In between, he co-founded FerretDB, a project that translates MongoDB queries into SQL for PostgreSQL or SQLite. MongoDB responded in May 2025 with a lawsuit alleging patent infringement, false advertising, and trademark misuse. FerretDB disputes all of it.
Asked what he'd learned from the experience, Farkas kept it short: "Lots of PTSD from courtrooms."
The FerretDB background also gave him an education in navigating open source communities, which turns out to be genuinely useful as Percona tries to be a credible player across multiple ecosystems rather than just the MySQL shop it started as. The PostgreSQL community, he discovered, does not respond well to the kind of vendor-branded product naming that MySQL users take for granted.
"If you call something 'Percona TDE for Postgres', that is not going to fly," he said. "'What is this nonsense? This is vendor lock-in!' Whereas in the MySQL space, that's totally normal." Percona had to unlearn assumptions baked in from two decades of MySQL-adjacent culture.
On whether Percona might eventually extend its open source ethos into AI itself, Farkas was sceptical but not dismissive. He pointed out that the dominant AI services right now are about as far from open source as you can get. "Frankly, AI is a vendor lock-in situation," he said, gesturing at the obvious examples.
He acknowledged there isn't yet a mature, enterprise-ready way to run open-weight agents, though Percona is apparently thinking about what it could contribute.
The closing line, though, is the one worth keeping: "Percona is not anti-AI. Percona is anti-rebranding Postgres or MongoDB as the AI database."
Someone had to say it.