AI deployments can fail for various reasons, but the key factor is often the architecture rather than the models or use cases. Simply adding an agent on top of an existing architecture may not work well, as classic architectures are not designed for agent workloads. Successful agentic deployments require a well-thought-out architectural framework with four key layers: the system of engagement, system of agency, system of work, and system of data.
To ensure a successful Agentic Enterprise, it is essential to follow eight design principles that focus on modularity, data connectivity, observability, trust, human intervention, event-driven processing, scalability, and open ecosystems. Designing for modularity means creating reusable components that can be easily integrated into different functions assigned to agents without rebuilding the entire architecture. Unified data and metadata are crucial for agents to make informed decisions and understand the context of their tasks.
Unified observability provides real-time visibility into agent actions, reasoning, context, governance, and business outcomes, allowing for optimization and troubleshooting when needed. Building trust into the architecture involves implementing governance and trust principles, such as distinct identities for agents and task-based permissions that expire after a specified period.
Strategic human intervention and oversight are crucial for ensuring that agents handle routine tasks efficiently and escalate to humans when necessary. Event-driven processing enables agents to respond in real-time to triggers across various channels, improving customer interactions. Scalability is essential for handling unpredictable AI workloads, ensuring that the architecture can adapt to spikes in computing needs without risking system failures.
Prioritizing open ecosystems and standards allows for interoperability and flexibility in integrating different models and applications. By following these design principles, companies can scale AI deployments effectively and build trust with regulators, customers, and stakeholders. The full report on these design principles can be found on Salesforce’s website for further reading and implementation.
For more Information, Refer to this article.
