Agentforce Specialist study resources

Trailhead is free, official and sufficient for the agent content. The section that catches people is Data 360 at 20% — candidates from an AI background consistently lose marks there rather than on prompt engineering. Budget study time by the published weights, not by what feels interesting.

Start with the official material

The official exam guide. Six sections: AI Agents 35%, Prompt Engineering 20%, Data 360 Fundamentals 20%, Testing Deployment and Maintenance 10%, Governance and Observability 10%, Multi-Agent Orchestration 5%.

Trailhead. Free, official, and the trails covering Agentforce and Data Cloud are written to this material. Work the agent trails thoroughly and do not skip the data ones.

Salesforce Help documentation for the security model specifics. The exam repeatedly tests that agents operate within the executing context’s permissions, and the documentation is more precise than any summary.

Get a developer org and build one agent

This matters more than any reading. A free developer org is enough to build an agent that:

• Handles a defined topic with a clear scope and an explicit out-of-scope refusal.
• Calls an action that writes a record, so you meet the difference between generated text and a deterministic, permission-aware operation.
• Requires confirmation before something consequential.
• Grounds a prompt template with merge fields from a related record, rather than typed values.
• Escalates to a human with the transcript and collected context preserved.

Build that once and the 35% agent section stops being memorisation.

The section people underestimate

Data 360, at 20%. Identity resolution across systems with different identifiers. Mapping source fields to a harmonised model. Streaming versus batch ingestion for continuously changing behavioural data. Why data quality matters more for an agent than a dashboard — because the agent acts on the data and asserts it to customers.

People from an AI background find prompt engineering easy and this section alien. It is worth the same as prompt engineering in marks.

Worth paying for

A practice test for diagnosis across six unevenly weighted sections. The free 20-question test here matches the real proportions.

Nothing else, if you already know Salesforce. Trailhead plus org time covers it.

Salesforce platform learning, if you do not know Salesforce. That is the real prerequisite and the harder one. Learning the platform first is faster overall than attempting this exam around the gap.

What to skip

Deep LLM theory. The exam does not ask how models work.

Braindumps.

Generic prompt engineering courses. The exam tests prompt templates grounded in CRM data with merge fields and defined output structure — a narrower, more concrete skill than general prompting advice.

A sensible order

1. Read the exam guide and write the six weights down.
2. Trailhead agent trails, then the Data Cloud trails. Do not stop before the data ones.
3. Build the agent described above in a developer org.
4. Take the practice test.
5. If Data 360 scored worst, go back there — it is the most common weak spot.

Where this sits

The Salesforce AI certification path covers who should and should not take this exam. If you build agents on another platform too, AB-620 vs Agentforce Specialist compares the two.