Agentforce Specialist cheat sheet
Last-minute reference for the Agentforce Specialist exam. The section people lose marks on is Data 360, not prompt engineering.
Exam facts
| Questions | 60 scored |
| Duration | 105 minutes |
| Passing score | 72% |
| Price | 200 USD |
A published percentage, unusually — so practice scoring is directly meaningful. Aim comfortably above it.
Section weights
| Section | Weight |
|---|---|
| AI Agents | 35% |
| Prompt Engineering | 20% |
| Data 360 Fundamentals | 20% |
| Testing, Deployment, and Maintenance | 10% |
| Governance and Observability | 10% |
| Multi-Agent Orchestration | 5% |
Agent constructs
| Construct | Purpose |
|---|---|
| Topic | What the agent handles; its classification description drives routing |
| Action | A deterministic, permission-aware operation — always preferred over generated text for a business operation |
| Scope | What is explicitly out of bounds, with a scripted refusal |
| Escalation | Transfer to a human with transcript and context preserved |
| Variable | Conversation-scoped to carry values between topics |
Data 360 essentials
- Identity resolution — match and merge records across systems into one profile
- Harmonised data model — map source fields to a standard structure
- Streaming vs batch — continuously changing behavioural data needs streaming
- Why quality matters more here — the agent acts on and asserts the data to customers
Recurring right answers
- Ground the template with merge fields from the related record, never typed values
- Specify the output structure explicitly when summaries come back inconsistent
- Require human confirmation before a refund or other consequential action
- Enforce entitlement at retrieval, before content reaches the model
- Deploy via sandbox to production, not by editing production
- Escalation and abandonment rates by topic are the useful production signal
Traps
- Agents cannot return data the executing context lacks rights to. The platform does not silently bypass sharing rules.
- Instructing a model to respect permissions is not a control.
- Salesforce platform knowledge is the real prerequisite, not AI knowledge.
- Data 360 is worth the same as prompt engineering — 20% each.
Night-before checklist
- Six sections and weights; AI Agents is 35%
- Passing score is 72%, a real target
- Security model: entitlement enforced at retrieval
- Check ID and proctoring rules — see exam day
Take the 20-question practice test.