Agentforce and certifications: what admins need to know
Agentforce is no longer a side topic in the Salesforce ecosystem: it has its own certification, and it has started appearing inside other exam outlines. Here is what it changes for administrators.
Two years ago, generative AI in Salesforce was a demo topic. Today it has an exam of its own, the Agentforce Specialist certification, and it has started to appear inside the outlines of certifications that have nothing to do with AI on paper. If you administer a Salesforce org, this is the shift worth understanding now.
The Agentforce Specialist exam in numbers
The exam follows the standard Salesforce format: 60 scored multiple-choice and multiple-select questions, up to five additional unscored questions, and 105 minutes on the clock. The detail that catches most candidates out is the passing score.
73% to pass. That is one of the highest bars in the entire Salesforce certification program. For comparison, the Administrator exam sits at 65%, Platform App Builder at 63%, and Data Architect at 58%. On a 60-question exam, 73% means you can afford roughly 16 wrong answers, not 21.
There are no formal prerequisites. You do not need the Administrator credential first, and you do not need to write Apex or Python. What you do need is hands-on time in an org with Agentforce actually enabled, because the questions are scenario-based rather than definitional.
What the five domains actually cover
The official outline splits into five domains, and the weighting tells you exactly where to spend your study time:
| Domain | Weight | What it really tests |
|---|---|---|
| AI Agents | 35% | Topics, actions, the reasoning engine, agent security, and connecting agents to channels |
| Prompt Engineering | 20% | Prompt Builder, template types, grounding techniques, and permission sets |
| Data Cloud for Agentforce | 20% | The Data Library, retrievers, and keyword vs. vector vs. hybrid search |
| Development Lifecycle | 20% | Testing Center, sandbox to production deployment, and adoption monitoring |
| Multi-agent Interoperability | 5% | MCP, the agent-to-agent protocol, and the Agent API |
AI Agents alone is worth more than a third of the exam. If you only have time to master one area, that is the one: how topics group actions, how the classification description drives intent recognition, and why an agent that works in a sandbox still has to be activated manually in production.
The part that surprises administrators
Three concepts account for a disproportionate share of wrong answers, and none of them are about AI in the abstract:
- The Agentforce Agent User is a real user. It executes actions, and it needs explicit permissions. An agent that cannot answer a Knowledge question is usually missing object access or the Data Cloud User permission set, not a better prompt.
- Activation is per environment. Testing an agent successfully in a sandbox tells you nothing about production. The agent, and any flow it calls, has to be activated there separately.
- The Einstein Trust Layer is exam material. Dynamic Grounding injects org data into the prompt, Data Masking is verified through Generative AI Audit Data, and you are expected to know which layer does what.
It is not just one exam anymore
The more interesting signal is what happened to Platform Developer I. Its refreshed outline now lists "understand the use cases and limits of using Agentforce for Developers" inside the Developer Fundamentals domain, which is worth 27% of that exam. The User Interface domain, another 25%, now explicitly mentions implementing Apex to work with Agentforce components.
In other words, Agentforce stopped being a specialty and became part of the platform baseline. The same thing happened to Lightning Web Components a few years ago: first a niche skill, then a specialist topic, then simply assumed.
What to do about it
If you are an administrator with no immediate plan to sit the Agentforce exam, the pragmatic move is still to learn the vocabulary. Topics, actions, grounding, and the Trust Layer are going to keep showing up in outlines you do care about, and in conversations with stakeholders who have read a press release.
If you do plan to sit it, treat the 73% bar seriously. Aim for consistent scores above 85% in timed practice before booking, because exam-day pressure and unfamiliar phrasing typically cost a few points against practice conditions.