A live AI agent, and the data pipeline behind it
A live AI agent that enrols and follows up with programme participants, and the Data Cloud ingestion that feeds it.
- Client
- Xero
- Role
- Associate Team Lead, Agentforce delivery team
- Through
- Cloudwerx
- Industry
- Software
How it fits together
Snowflake reconciliation table
New source added by the client
Data Cloud data stream
New ingestion
Data model object
Mapped to the same object as the original source
Triggered Flow updates Accounts
Existing automation, unchanged
Agentforce agent
Topics, actions, prompt templates, Apex and Flow
Orange marks the parts I built.
Context
Xero's Challenge agent is a live Agentforce agent. It identifies eligible programme participants, checks consent and opt-out status before enrolling anyone, and drives completion through automated email outreach and follow-up.
What I built
- Agent topics and actions
- The instructions that decide what the agent handles and the actions it can take.
- Prompt templates
- Written and tested in Prompt Builder.
- Supporting Apex and Flow
- The same bulk-safe, idempotent patterns I use on enterprise integrations, so agent-driven outreach stays reliable at volume.
- Testing and tuning
- Checking and adjusting how the agent behaves.
Data Cloud
When the client introduced a new reconciliation table in Snowflake, I built a new data stream ingestion and mapped it to the same data model object as the original source. The team's existing Data Cloud-triggered Flow picks up the reconciled data and creates or updates Account records, with no change needed downstream.
Outcome
Live. I have been working on both the Agentforce and the Data Cloud side since August 2026.
Skills used
- Agentforce
- Prompt Builder
- Data Cloud
- Snowflake ingestion
- Data model objects
- Flow
- Apex
Next case study
A reusable approval and special-conditions framework