Native Salesforce AI:
- Einstein AI — embedded scoring, predictions, classifications.
- Agentforce — autonomous AI agents.
- Atlas Reasoning Engine — Salesforce's LLM platform.
- Prompt Builder — reusable prompt templates.
External AI:
- OpenAI / Anthropic / Google APIs called from Apex.
- AWS Bedrock / SageMaker for AWS-hosted models.
- Azure OpenAI for Azure ecosystems.
Patterns:
1. Generative actions.
- User triggers; LLM generates (email draft, summary, recommendation).
- Apex calls LLM via Connect API or external API.
2. Classification.
- Records auto-classified (case category, lead score).
- Trained on historical Salesforce data.
3. Conversational interfaces.
- Chatbots (Einstein Bots) handle initial contact.
- Hand off to human when needed.
4. Agent assist.
- AI suggests responses to agents in real time.
- Agent reviews and accepts.
5. Document understanding.
- AI extracts data from uploaded docs.
- Populate Salesforce records automatically.
Architecture decisions:
- Trust Layer mandatory for sensitive data.
- Fallback paths when AI unavailable.
- Cost monitoring per call.
- Human-in-the-loop for high-stakes.
- Audit every AI decision.