Scheduling: AI assistant
Objective. Pose a scoped scheduling question or change in natural language and inspect what the assistant proposes and the rules and forecast it used - before trusting or publishing it.
The assistant answers: can I ask for a schedule change or explanation in plain language? The key read is that a fluent proposal is still a drafted suggestion bound by the same rules and forecast, and must be verified against its evidence, not trusted for its wording.
See it in the demo
What this screen answers
A Sales or CSM person uses the assistant to answer: can I get help building or explaining a schedule by asking in natural language - for example 'cover the 2pm shortage without overtime'? It lowers the barrier to driving the planner and explaining what a view means.
The task
Pose a scoped, concrete request or question for the team and period. Inspect what it proposes: which intervals it changes, which rules and skills it respected, and which forecast it worked from. Treat the output as a draft action to review, not an instruction to follow blindly.
How to read it
The assistant proposes against the same rules and forecast as the rest of Scheduling; its suggestion is a draft action, not an executed or published change. Fluent, confident wording is not proof of correctness. Inspect the evidence - the intervals affected, the rules honoured, the forecast version used - before accepting. The misread to avoid is treating a well-phrased answer as a done or verified change.
Honest limits
The assistant cannot be more correct than the forecast and rules behind it, and its proposals still have to be published deliberately to become live. Confirm with product what the assistant is actually entitled to change versus only suggest for a customer; the demo runs on seeded data.
Interpretation check
An assistant proposal is a drafted suggestion bound by the same rules and forecast - verify its evidence and publish deliberately; fluent wording is not proof of correctness.
In the field
A new scheduler asking the assistant to cover a peak without overtime: useful to draft options fast, but the CSM still verifies the proposal's evidence before it is published.
Practical task
On the demo, ask the assistant to address one shortage window, then write a one-line verification: which intervals/rules/forecast its proposal used and whether you would publish it.
Self-check
The assistant gives a confident, well-worded schedule change. Can you trust it?
Not on wording alone. Inspect which intervals it changed, which rules it respected and which forecast it used. Fluency is not correctness.
Does accepting an assistant suggestion make the change live?
No. It is a draft action. It becomes live only when published deliberately, on the correct forecast with violations cleared.
What limits the quality of any assistant proposal?
The same thing that limits the whole app: the forecast version and the configured rules. It cannot be more correct than its inputs.
Verified vs demo. Screen is live in the demo at /#/assistant; exact controls, labels and figures are SME-validation-pending.