What managing 150,000 AI agents could look like for database teams

The database admin job is heading for a weird twist: less direct database babysitting, more supervision of software that does the babysitting for you. That sounds backwards until you look at what AI agents are starting to do. They can inspect a problem, decide what needs attention, use tools to act, and check whether the fix worked. In other words, some of the work DBAs have spent decades doing by hand is moving one layer up.

That doesn’t erase the human role. Someone still has to define what an agent is allowed to touch, when it needs approval, and what happens when it gets things wrong. But the center of gravity shifts. A DBA dealing with slow queries or provisioning no longer has to handle every step personally. The job becomes more about setting rules and watching the systems that carry them out.

The scale problem is what makes this real, and a little alarming. Gartner says the average global Fortune 500 company will have more than 150,000 AI agents in use by 2028, up from fewer than 15 in 2025. Only 13% of organizations think they have the right governance for that. Not every agent needs its own database, but many will create state, retrieve data, remember prior interactions, and talk to other agents. And they won’t behave like normal applications. They may sit idle for long stretches, then suddenly get busy.

Yugabyte is aiming at that mess with YugabyteDB AMP, which stands for Agentic Multitenant PostgreSQL. The idea is to pack hundreds of small Postgres workloads onto shared distributed infrastructure while keeping the databases isolated. Lifecycle tasks like provisioning, branching, scaling, migration, and teardown can be exposed to agents through MCP, and Yugabyte has built agents for setup, migration, performance tuning, and integrations. The pitch is simple: don’t provision database number 14,372 like it’s special when it isn’t.

The economics matter too. AMP uses serverless multitenancy and scale-to-zero, so workloads can share infrastructure and idle agents consume no compute. Customers pay by CPU minute, and resource limits can keep one overeager agent from eating everyone else’s lunch. Yugabyte also wants the same stack to cover the full journey, from cheap experiments on serverless Postgres to fully distributed YugabyteDB when something gets serious. Above that sits Meko, its agent-native context engine, which keeps shared memory, decision traces, and auditable context across agents instead of trapping everything inside one silo.