afterword
If the last decade was about making analytics self-serve, the next decade will be about making analytics self-operating. That sounds like a small shift in wording, but it isn’t. “Self-serve” assumes humans are still the control plane: we notice something’s off, we investigate, we patch the pipeline, we add a guardrail, we write a postmortem, we remember the context. “Self-operating” means a meaningful portion of that work moves to agents—systems that can observe your data platform, propose changes, execute workflows, validate outcomes, and keep iterating.
This is the part that feels visionary, because it changes what your analytics system is. It’s no longer just a place where people ask questions. It becomes an environment where software takes actions.
And here’s the inconvenient truth: most analytics stacks today are optimized for answering questions, not for taking responsibility. They return results. They don’t guarantee that those results are reproducible next week, explainable next quarter, or safe to change next year. Humans have been the error-correcting layer. Agents won’t be, unless you build the system so that “careful” is the default.
In practice, agents that manage data need something closer to database semantics than file semantics. They need a platform where data is treated as state, not just blobs in object storage. They need to be able to say, with confidence,