EstarWorks · Phase one of deployment
Before the agent, the boundaries.
An AI agent is only as safe as the limits it works inside. Before any agent touches your business, we sit down with you and define, in writing, three things: what data it may access, which actions it may take on its own, and which actions it must stop and put in front of a person for approval. Those boundaries are mapped to your real workflows — not a generic policy — and they are built into the system, not left as good intentions.
The three zones
Acts on its own
Routine, reversible work inside clear rules:
- Answering standard questions
- Capturing enquiry details
- Booking within your calendar rules
- Confirmations and reminders
Prepares — a person approves
Anything that commits the business. The agent drafts and assembles; a named person decides:
- Refunds and discounts
- Payments and invoices
- Contract terms
- Outbound messages outside approved templates
Never touches
Data and decisions outside its brief, by design:
- Banking, credentials and payment details
- Bulk export of client lists
- Personnel matters
- Anything you mark off-limits
If something is not on the agent's access list, the agent cannot reach it. Access is granted field by field and system by system, and the list is a written document you hold.
Mapped to your workflows
Boundaries only work when they match how your business actually runs. We work through the places an agent will meet real decisions:
- Client lists and CRM records — which fields the agent may see, search and update, and what it may never export.
- Payments and invoices — the agent may prepare, send and chase; moving money stays with your people.
- Refunds — eligibility checked against your own written rules; approval and execution stay with a person.
- Outbound messages — approved templates and tone are set in advance; anything unusual is drafted for review, not sent.
- Quotes and pricing — prepared from your price book; anything beyond your stated limits needs a person's approval.
Why owners ask for this first
The real question is not "is the AI smart enough?"
It is "what could it touch, and who says yes?" Every large platform ships its own general safeguards, but your business's real boundaries are specific: which client list never leaves the building, which payment needs two pairs of eyes. Nobody defines those for you — they are defined here, with you, before anything is connected.
How it fits
Boundary and approval design is phase one of every AI staff deployment we do. It happens before the build, as part of the diagnosis and roadmap, and it is revisited whenever your workflows change. It is not a separate product — it is how every deployment starts. Deployment itself is a one-time engagement from CA$1,999; the full path is on the AI Staff Deployment page and the pricing page.