Most water utilities already have a dashboard. Many are now being offered an agent. The two are usually presented as successive generations of the same thing, which they are not: they answer different classes of question, and choosing between them is a question about your questions rather than about the technology.
What a dashboard is good at
A dashboard encodes a fixed set of questions as a fixed set of views. Reservoir levels by zone. Non revenue water by district. Pressure against the standard. Once built, it answers those questions instantly, identically, for everyone, forever, and it is auditable: the number on the screen came from a query somebody can read.
That is a real strength for a standing question asked daily by many people. It is also the reason dashboards fail: the question a team actually has at 2am during an event is rarely one of the twelve the dashboard was scoped around. We wrote separately about why water dashboards go unused, and the pattern is almost always this one.
What an agent is good at
An agent takes a question in plain language, decides which data and which tools it needs, runs them, and returns an answer. The value is not the language interface. It is that the set of answerable questions is no longer fixed at build time.
"Which DMAs had night flow above their baseline for three consecutive nights last month, and what was the rainfall in those zones" is a question no one scopes into a dashboard, and an analyst would take a morning to answer. That is the shape of question an agent earns its place on.
Our own GeoPilot is built on this premise for GIS work: run real geospatial analysis from plain language, so the work does not stall waiting on a specialist.
Where the agent is the wrong answer
Three cases, and they are common.
- A regulatory number. If a figure goes into a statutory return, it needs one definition, computed one way, traceable. That is a query, not a conversation. Non determinism is a defect here.
- A question asked a hundred times a day. If everyone asks the same thing, build the view. An agent that re-derives it each time is slower and more expensive.
- Data that is not ready. An agent over inconsistent, uncalibrated or partly missing data produces fluent answers that are wrong, and fluency makes them harder to challenge than a blank dashboard panel.
The honest comparison
The choice tracks question variety, not sophistication. Few questions asked often: dashboard. Many questions asked rarely, by people who can judge the answer: agent. Most utilities need both, with the dashboard carrying the daily operational picture and the agent carrying investigation.
What decides whether either works is the layer underneath. Both fail on the same thing: data that has not been reconciled, and no model of how the water behaves. That is the argument for treating this as a decision support system rather than an interface project.