Agent Scope Contract (Onboarding)

Agent Scope Contract (Onboarding)

1. 01

   Name the single client workflow the agent will own and write it as one sentence

   Reject any brief that describes the agent as a general assistant. A named workflow such as inbound lead triage for a 40-person B2B sales team gives you a testable boundary; a vague one gives you an open-ended retainer.

2. 02

   Label the engagement stage before scoping: proof of concept, prototype, pilot, or MVP

   Each stage answers a different question and carries a different deliverable. Skipping the pilot stage is the most common agency error when placing an agent into a client environment, and it is where scope disputes start.

3. 03

   Map every tool call and data source the agent will touch, with the client owner named for each

   List the CRM, helpdesk, calendar, and file stores explicitly. For each, record who at the client approves read and write access, so permission delays do not land on your delivery timeline.

4. 04

   Set the control tier: what the agent may do autonomously, what needs human approval, and what is blocked

   Write the approval gate into the scope document, not the build notes. Agent-generated output that reaches a client dashboard without a review checkpoint is the failure mode that erodes trust fastest.

5. 05

   Choose the build surface against the workflow, not against the demo

   A white-label shell suits agencies reselling a branded client-facing assistant; a typed, versioned behavior layer suits workflows that need audit trails; a local-first runtime suits clients with data residency constraints. Match the surface to the constraint before pricing.

6. 06

   Document who controls metering, feature access, and dispute resolution for each vendor in the stack

   Vendors that set usage limits, gate features, and arbitrate their own disputes concentrate risk on your delivery. Flag any single vendor holding all three roles and price a fallback path into the proposal.

7. 07

   Price the integration, governance, and testing work as line items separate from the agent shell

   The shell is the cheap part. Itemize knowledge ingestion, evaluation harness, monitoring, and handover documentation so the client sees where the defensible value sits and your margin survives renewal.

8. 08

   Get written client sign-off on scope, control tier, and stage label before any build starts

   A signed one-page scope beats a 30-slide deck. It becomes the reference point for every change request and the boundary you cite when the client asks for a second workflow at no cost.