Agent Scope Contract (Onboarding)

Agent Scope Contract (Onboarding)

1. 01

   Name the single client workflow the agent will own before any builder is opened

   Write it as one sentence with a measurable output, for example 'qualify inbound demo requests and route to the right rep within 5 minutes'. Multi-workflow agents are the most common cause of scope creep in month two.

2. 02

   Map every tool call the agent needs and mark each as read or write

   A builder that can post to Slack, update HubSpot, or send email carries different liability than one that only reads a knowledge base. Record the credential owner for each connection.

3. 03

   Set the autonomy ceiling and the human checkpoint for anything client-facing

   Classify the workflow as draft-only, approve-before-send, or fully autonomous. Any step touching outbound communications or CRM records gets a named reviewer and a response window.

4. 04

   Define the memory and context boundary in writing

   State what the agent may retain between sessions, where that data lives, and whether client data enters a shared model pool. Private deployment choices change both pricing and the client conversation.

5. 05

   List the acceptance tests the client will run at handover

   Pick 10 to 20 real historical inputs with known correct outcomes. If the agent cannot clear that set, the build is not ready for a retainer conversation.

6. 06

   Price the build, the run cost, and the maintenance retainer as three separate lines

   Build fees recover setup hours, usage fees pass through model and infrastructure cost, and the retainer covers prompt drift, tool breakage, and monthly review. Bundling all three hides margin erosion.

7. 07

   Get written sign-off on scope, autonomy level, and data handling before build starts

   A one-page signed scope prevents the mid-project request to 'also handle support tickets' from becoming unpaid delivery work.