Daily workflow
AI agents and automation
Put scoped AI agents to work answering operational questions and drafting changes a human approves.
Owner: ManagerAgentsApprovalsScheduled tasksAI tokens
Walkthrough video
AI agents and automation walkthrough
A short product video or annotated screenshot showing this workflow from the operator's point of view.
What happens
Follow this routine when your team is doing this work for real.
- 1Open Agents and create an agent with a name, persona, and instructions that describe its job, such as a dispatch assistant or a payroll-exception reviewer.
- 2Choose the model (Haiku, Sonnet, or Opus) — cheaper is faster, pricier reasons better — and set a monthly budget so spend stays predictable.
- 3Grant access per module: pick exactly which areas (jobs, schedule, customers, resources, inventory, leave, payroll) the agent may read, and which it may propose changes to.
- 4Chat with the agent for one-off questions, or create a scheduled task so it runs on a cron schedule (for example, every Monday 9am summarize last week's payroll exceptions).
- 5Leave the agent in Ask mode so every write becomes a proposal; switch trusted, low-risk tasks to Auto mode only when you want changes applied immediately.
- 6Review everything waiting in the approval inbox on the Agents page — approve or reject each proposed change in one place.
What this uses
The records Flowdexa connects behind the scenes.
- Permissions
- An agent can only read and propose changes to the modules you explicitly allow, scoped to the one project — it never reaches across projects or beyond its grants.
- Proposals
- In Ask mode every write is a proposal that lands in the approval inbox; nothing changes operational data until a person approves it.
- Scheduled tasks
- Tasks run a saved prompt on a schedule. They fire automatically when deployed, and can be fired on demand with Run due now.
- AI tokens
- Each run consumes AI token credit against the agent's monthly budget and the project's credit balance, billed per model usage.
- Audit log
- Every agent read, proposal, and applied change is recorded in the audit log under an agent actor, so activity is fully traceable.
Before you call it done
Run these checks to make sure the workflow is actually working end-to-end.
- The responsible role can complete the workflow without owner access.
- Required customer, service, schedule, resource, or payroll records are connected.
- Blocked work has a visible reason and a named next owner.
- Managers can see the status in daily operations or reports.
Next
Keep learning daily workflows
Use the workflow index when training another role or moving to the next part of daily operations.