Schedule Agent

The Dispatch Board is a calendar view of work orders across technicians. Dispatchers juggle skills, travel, overtime, and shifting priorities manually. Schedule Agent generates an optimized schedule from thousands of combinations and presents it as a draft to review and commit.

The Problem

No holistic optimization. Priority, skill matching, location, and travel aren't coordinated as volume changes. The schedule degrades quickly.

Decisions

The agent drafts, the dispatcher commits. It could write straight to the board, and for a scheduling problem this size that would be the efficient choice. It does not, because the dispatcher holds knowledge the constraints never capture, and a schedule they did not approve is one they will quietly work around.

Constraints are the interface. There is no screen for the optimization itself, so the design went into how a dispatcher expresses what matters: hard rules and soft preferences, each editable in plain language, saved as profiles, and compared side by side. A request can carry more than one profile, so the choice is between drafts, not between the agent and nothing.

Review happens on the board people already know. The draft shows what changed, what stayed, and what was left out on the same calendar, rather than in a report. Validation runs again at commit, since the world moves while the engine runs.

Value is proven against a baseline, not a memory. Once a team runs the agent every week there is no manual schedule left to compare with, so the engine simulates one. A generous baseline that still loses is the only comparison a skeptical manager will accept.

Results

Measured against the shadow baseline, the agent schedules more work with less travel and fewer missed SLAs. Teams with more work than hours gain throughput. Teams with slack gain efficiency, less driving and less overtime.