What did Google Cloud introduce in August to support AI agent infrastructure—specifically, how do its Gemini Cloud Assist–integrated Database Operations Agents (including the natural language Onboarding Agent and telemetry driven Observability Agent, their supported database serv

In August, Google Cloud introduced two complementary building blocks for always-on AI agents: Database Operations Agents to automate database setup and operations, and Cloud Run instances for low-cost, persistent single-agent runtimes. Together they extend Google’s Agentic Data Cloud strategy from data management into the operational infrastructure agents need. 78

Database Operations Agents

• The Gemini Cloud Assist–integrated offering comprises a Database Onboarding Agent for Day 0 work—database selection, setup, configuration, and initial deployment—and a Database Observability Agent for Day 1/2 monitoring, troubleshooting, and maintenance. 8

• The goal is to replace repeated manual architecture planning, scripts, provisioning, schema/index configuration, query tuning, and diagnostics with natural-language, guided workflows—reducing DBA, developer, SRE, and DevOps toil. 8

• The Onboarding Agent guides users through database provisioning and initial configuration in natural language, supporting Google Cloud managed databases including AlloyDB, Bigtable, Cloud SQL, and Spanner. 8

• The Observability Agent correlates telemetry from Database Insights, Cloud Monitoring, Cloud Logging, Cloud Trace, and related sources to identify root causes such as latency spikes, query hot spots, lock contention, or resource constraints. 8

• It is available across operational interfaces: Gemini Cloud Assist chat, relevant Google Cloud console database pages, Database Center fleet views, the CLI, IDEs and third-party tools. 8

• Its Database Insights and Database Center capabilities are also exposed as MCP tools through Google Remote MCP servers, so external developer tools and agents can retrieve system/query metrics, fleet inventory, and detected issues. 8

• The agent does more than diagnose: it proposes validated changes with rationale and expected impact, and—subject to user approval—can execute supported remediations, such as enabling Cloud SQL connection pooling or adding an index. Some investigation and remediation functions were in preview for selected customers. 8

Cloud Run instances

• Cloud Run instances, introduced in preview in late August, are dedicated singleton Cloud Run runtimes intended for long-lived, stateful, typically single-user workloads such as personal AI agents. 9

• Unlike ordinary Cloud Run services, an instance runs exactly one copy, does not autoscale or scale to zero, can run continuously for up to seven days with an automatic restart policy, and may be stopped and resumed. 9

• Each instance receives a stable HTTPS URL that remains unchanged through updates and restarts. 9

• Google positions this as an alternative to a continuously operated VM: it avoids VM operating-system maintenance, firewall and HTTPS-endpoint setup, while retaining a persistent process model that conventional scale-to-zero serverless services do not provide. 9

• The stated continuous 30-day price for 1 shared vCPU and 1 GiB memory is $5.70, using burst budgets for occasional compute spikes rather than sizing and paying for a full VM all the time. 9

• Persistent execution also reduces cold-start exposure; one reported early user said its long-running agent saw cold starts fall 88%. This is a customer claim, not a general performance guarantee. 9

Strategic significance

Google is addressing both sides of agent infrastructure: agents need durable, economical execution and stateful interaction, while the databases they depend on need automated provisioning, observability, and safe remediation. Cloud Run supports agents that orchestrate asynchronous work and multi-turn interactions, while the database agents make the data layer increasingly operable through natural language and MCP-compatible tools. 18

A practical limitation remains: “stateful” execution does not eliminate the need for durable external state if an agent must survive restarts or preserve work beyond a runtime’s lifetime; Google’s guidance also points to persistent session storage such as Cloud SQL for production agents. 5