Einstein and AI features in Salesforce
Einstein still names many Salesforce AI features and the Trust Layer. Here is how they fit with Agentforce today.
Einstein was Salesforce's original brand for AI across its clouds. The agent experience has since moved under Agentforce: Einstein Copilot became Agentforce Assistant. The Einstein name has not disappeared, though. It still labels specific features, the Einstein Trust Layer, and the Setup step where "Einstein and Agentforce must be enabled".
What it is for
Salesforce AI features fall into two broad groups. Predictive features score, classify or recommend based on patterns in your data. Generative features draft text, summarise records and power agents. Both depend on good data and clear permissions.
Documented AI features and their status
| Feature | What it does | Status per Salesforce |
|---|---|---|
| Einstein Trust Layer | Grounding, data masking, toxicity detection, audit trail, zero data retention with LLM providers | Available |
| Einstein Conversation Insights | Analyses sales calls and meetings | Available |
| Prompt Builder reference visibility | Shows what a prompt references | GA at Dreamforce 2026 |
| Prompt Builder conditional expressions and batch prompt testing | Richer prompt logic and testing | Beta |
| AI-drafted knowledge articles | Drafts Service Cloud knowledge (needs a Knowledge add-on licence) | Winter '27 |
| Koa | Salesforce's first CRM reasoning model | Pilot; GA expected winter 2026 in U.S. regions |
The Einstein Trust Layer
The Trust Layer is the set of controls between your data and the large language models. It grounds prompts in CRM data, masks sensitive data, checks for toxic output, keeps an audit trail with feedback, and relies on zero data retention agreements with LLM providers. Developer docs and Help call it the Einstein Trust Layer; some 2026 newsroom copy says Agentforce Trust Layer. Both refer to the same idea.
How it connects with other Salesforce products
AI features draw on CRM data and, where configured, Data 360. Sales Cloud uses Einstein Conversation Insights, Service Cloud offers AI-drafted knowledge, and Marketing Cloud Next uses agents such as the Agentforce Content Agent.
Who it suits
Any Salesforce customer that wants AI help inside existing processes, starting with a small, measurable use case.
Planning AI adoption
- 01
Check licences
Confirm which AI features your editions include.
- 02
Enable carefully
Turn on Einstein and Agentforce in a sandbox first.
- 03
Review data access
Make sure permissions and sharing reflect who should see what.
- 04
Test prompts
Test with real records and review outputs before rollout.
- 05
Measure
Track adoption and quality, not just usage.
Common mistakes
- Assuming every Einstein feature is included in every edition.
- Enabling generative features before fixing record access.
- Treating beta or pilot features as production-ready.
How GeekMidst can help
Our Agentforce and AI automation service can help you choose and test AI features. Our org health check reviews data and permissions first, and our responsible AI page explains how we work.
Questions we hear
Has Salesforce retired Einstein?
Is it the Einstein Trust Layer or the Agentforce Trust Layer?
Does Salesforce train models on my data?
Which predictive features do I have?
Related reading
Agentforce guide
The agent platform explained
→Security and data handling
How we handle your data
→Writing buildable requirements
Turning AI ideas into clear requirements
→Sources
Exploring Salesforce AI?
Tell us which processes you want to improve, and we will help you choose and test the right features.
