The next database challenge won't be handling massive single datasets, but managing hundreds or thousands of smaller databases that AI agents create and abandon. That's the premise behind YugabyteDB AMP, or Agentic Multitenant Postgres, a serverless PostgreSQL offering unveiled by Yugabyte to handle what co-founder and co-CEO Karthik Ranganathan calls a shift in the dimension of scale. The platform uses serverless multitenancy and Enhanced Colocation to pack hundreds of workloads onto shared infrastructure while giving each agent an isolated PostgreSQL database that scales to zero when idle and charges by CPU minute when active.
YugabyteDB AMP addresses the economics of agent-driven database sprawl, where an experimental agent might heavily use its database for a few minutes before sitting untouched for days or being destroyed entirely after completing a task. The platform ships with four built-in AI agents: Architect handles database creation and provisioning while retaining organizational rules about where development and staging databases should live and how much compute they receive. Voyager manages migration and modernization, working out sizing and scaling requirements for bringing existing databases onto PostgreSQL. Perf Advisor investigates outages or performance problems and attempts to determine root causes, with plans to warn developers when queries might cause trouble before reaching production. Nexus connects databases to warehouses, data streams, key management systems, and other external infrastructure. These agents share context about the organization and maintain decision traces that let users reconstruct why particular actions were taken.
According to Ranganathan, the agents operate with what he describes as a "guarded amount of freedom," with autonomy levels depending on the environment. A disposable development database might let the agent proceed without repeatedly asking permission, while a production system holding financial information would use dry run mode where the agent explains what it wants to do before taking action. The platform is designed so developers don't have to switch databases when workloads grow from experimental to business-critical. As demand increases, YugabyteDB can shift to YSQL, its distributed PostgreSQL layer, adding horizontal write scaling, multi-master replication, and geo-distribution without requiring an application rewrite or conventional migration. The company also offers a bring-your-own-cloud deployment option where AMP operates inside a customer's own cloud environment while connecting back to YugabyteDB's control plane for metering, management, and insights.
YugabyteDB AMP isn't broadly available yet, with Yugabyte currently using it internally and with a small number of selected partners before wider rollout. The company is already running agents internally for coding, support, and go-to-market work, and is moving other internal databases onto the platform to consolidate its own database sprawl. Yugabyte's bet doesn't require every AI agent to survive, but rather assumes companies will create many agents where most spend time idle, some disappear, and a small number unexpectedly become critical. If that scenario plays out, the database problem shifts from finding somewhere to put one enormous workload to keeping track of many small ones that humans can't reasonably manage alone.