Salesforce Agentic Enterprise Design Principles
A framework for implementing AI agent architectures in enterprise environments, emphasising modularity, data harmonisation, observability, trust, and scalable infrastructure.
-
Design for modularity
Build reusable, composable components that can be reconfigured for different agent functions without rebuilding entire architecture elements from scratch.
-
Harmonise data with metadata-driven understanding
Provide agents with unified data and metadata, business glossaries, and ontologies so they can understand context and make informed decisions across systems.
-
Enable unified observability
Maintain real-time visibility into agent actions, reasoning, context, governance, and business outcomes across both IT and business functions.
-
Build with trust
Embed governance and trust principles throughout the architecture, including agent identities, task-based permissions, and policy compliance checks.
-
Design for strategic human intervention and oversight
Implement appropriate human-in-the-loop engagement at decision points, with smooth handoffs that preserve context when escalation is needed.
-
Enable event-driven processing
Support real-time decision-making through always-on agents that respond to triggers across any channel, including texts, emails, calls, and APIs.
-
Ensure infrastructure can scale growing AI workloads
Build resilient infrastructure with compute that scales based on workload needs, distributed loads, and storage that handles unpredictable data access patterns.
-
Prioritise open ecosystems and standards
Design for interoperability using standard interfaces, open protocols, and portable workflow definitions to avoid vendor lock-in and enable flexibility.