Databricks has published a primer on a term Gartner coined in 2017, and it explains why it is writing one now. In What is AIOps?, the answer is that “the infrastructure running your applications in 2026 looks nothing like what traditional monitoring was built for”: hundreds of microservices, multi-cloud dependencies and “agent-driven systems that generate more operational signals in an hour than an on-call engineer can read in a week.” AIOps, as it defines it, applies AI and machine learning to detect anomalies, correlate events, find root causes and trigger responses, and it “complements observability and DevOps rather than replacing them.”
There are no figures, and the post is careful in a way vendor primers often are not. It says what AIOps does not do: “It does not replace observability, DevOps, or human judgment.” Its walkthrough is the right one: a spike in API errors is compared against history, correlated with a rise in database load and a recent deployment, and attributed to the deployment rather than treated as three incidents. The final stage is “guided or automated response,” restarting a service, scaling infrastructure, rolling back a deployment, with one rule stated plainly: “Higher-risk actions should require human-in-the-loop to allow engineers to review and approve the response before it affects production.” For agent-built remediation it points to Agent Bricks.
The catalog holds a strong record for the data side of this. The Databricks provider page lists API pages across the platform, and an AIOps pipeline as the post describes it ingests into the lakehouse through the Databricks Pipelines API, queries through the Databricks SQL Warehouses API, serves its models through the Databricks Serving Endpoints API, and governs who may act through the Databricks Permissions API. On the Agent Readiness score, idempotency, reversibility, error semantics, OpenAPI examples, the event surface and agent skills are all lit, which is a better operational base than most platforms this series covers.
The Kin Score is 67.6, exemplar band, carried by developer ergonomics at 88.6 and operational transparency at 63.2, with contract governance at 43.6. The Agent Readiness score is 53.5, agent-ready. Two unlit dimensions meet the post’s own rule. Dry-run mode is unlit, and it is the contract-level version of “review and approve the response before it affects production”: a way for an agent to say what it would do before doing it. The MCP server is unlit too. Databricks has written the human-in-the-loop principle into its definition of AIOps. The contract an AIOps agent would act through has reversibility and idempotency, and does not yet have a rehearsal.