Enhancing Enterprise Operations Through Multi-Agent Collaboration: A Case Study with Salesforce Agentforce and AWS
Overview of Multi-Agent Collaboration in Enterprise AI
Introduction to Agentforce
Key Features of Amazon Bedrock Agents and Knowledge Bases
Integration Patterns Between Agentforce and Amazon Bedrock
Detailed Solution Overview
Prerequisites for Implementation
Preparing Amazon Redshift Data
Steps to Create IAM Roles
Creating an Amazon Bedrock Knowledge Base
Creating an Amazon Bedrock Agent
Developing a Lambda Function for API Integration
Setting Up a REST API in API Gateway
Configuring Named Credentials in Salesforce
Adding an External Service in Salesforce
Creating an Agentforce Action for the External Service
Configuring the Agentforce Agent to Utilize the Action
Clean-Up Procedures
Conclusion: The Future of AI-Driven Workflows in Enterprises
About the Authors
Transforming Enterprise Workflows with Multi-Agent Collaboration
In today’s rapidly evolving business landscape, AI agents are revolutionizing the way enterprises operate. While individual agents excel at executing specific tasks, complex workflows often traverse multiple systems, necessitating collaboration across various platforms. This is where multi-agent collaboration comes into play — allowing specialized AI agents to work together, streamlining intricate business processes.
In this blog post, we’ll delve into a practical example of such collaboration, integrating Salesforce Agentforce with Amazon Bedrock Agents and Amazon Redshift to automate enterprise workflows.
Multi-Agent Collaboration in Enterprise AI
Modern enterprise environments are intricate, populated with diverse technologies and numerous systems. Salesforce and AWS each offer unique advantages, and many organizations have substantial infrastructure already built on AWS. This includes a plethora of data, AI services, and business applications — from ERP and finance to HRMS and supply chain management.
Agentforce provides powerful AI-driven capabilities that harness enterprise-specific data, but to truly maximize potential, organizations need agents that can interact and act on information across various platforms. By integrating AWS-powered AI services into Agentforce, businesses can orchestrate intelligent agents that operate seamlessly between Salesforce and AWS, merging their respective strengths.
Let’s explore the collaboration models between Agentforce and Amazon Bedrock Agents:
- Agentforce as the Primary Orchestrator:
- Manages end-to-end customer-oriented workflows.
- Delegates specialized tasks to Amazon Bedrock Agents using custom actions.
- Coordinates access to external data and services across systems.
This integration results in a robust solution, enhancing business outcomes through advanced AI capabilities and seamless cross-system functionality.
Agentforce Overview
Agentforce brings digital labor into every aspect of business, empowering employees and enhancing customer experiences. It integrates effortlessly with existing applications, data, and business logic, facilitating meaningful action across the enterprise. Key features include:
- Deployment of pre-built agents tailored for specific roles and industries.
- Agents that can act on existing workflows, code, and APIs.
- Secure connections to enterprise data.
- Delivery of accurate, grounded outcomes via the Atlas Reasoning Engine.
Amazon Bedrock Agents and Knowledge Bases
Amazon Bedrock is a fully managed service that provides access to high-performance foundation models (FMs) through a single API. Key features include:
-
Amazon Bedrock Agents: These managed AI agents leverage FMs to understand user requests, break down complex tasks, and orchestrate actions. They can interact with various company systems via APIs and access information using knowledge bases.
-
Amazon Bedrock Knowledge Bases: This feature enables agents to perform Retrieval Augmented Generation (RAG) using private data sources. When tasked with a query, the agent can retrieve relevant information from a connected knowledge base, delivering precise, context-aware responses without requiring retraining.
Agentforce and Amazon Bedrock Integration Patterns
Agentforce can interact with Amazon Bedrock Agents in various ways, allowing for creative architecture solutions. Common interaction patterns include:
-
Synchronous Interactions: Utilizing custom agent actions and recorded integrations, Agentforce can relay requests to Amazon Bedrock through secure means, like using named credentials to authenticate securely.
-
Asynchronous Interactions: Implementing Salesforce Event Relay alongside Amazon EventBridge to communicate with Amazon Bedrock Agents.
In this post, we will primarily focus on synchronous interactions, highlighting how to build interactive workflows.
Solution Overview
To demonstrate the power of multi-agent collaboration, we’ll outline a workflow that uses IoT sensor data. For instance, let’s take Coral Cloud, which has equipped its facilities with smart air conditioners and temperature sensors. The workflow processes the following:
-
Data Capture and Storage: Sensors record data like temperature and error codes and store it in AWS’s Amazon Redshift.
-
Query Execution: When Agentforce queries an Amazon Bedrock Agent asking, “What is the temperature in room 123?” it does so through an API call powered by AWS Lambda.
-
Contextual Processing: The Amazon Bedrock Agent retrieves necessary contextual information from its connected knowledge base, deciding on the appropriate action based on the query and data received.
-
Error Handling: If sensor readings indicate errors, the Amazon Bedrock Agent creates a case in Agentforce, utilizing the Agent Wrapper Lambda function for direct interaction with Salesforce.
This architectural flow demonstrates how Amazon Bedrock Agents can execute tasks, leverage contextual insights, and utilize action groups to interact with Agentforce, ensuring a seamless end-to-end process.
Prerequisites for Building This Architecture
To build this integration, ensure you have:
- An active AWS account with permissions for Amazon Bedrock, Lambda, Amazon Redshift, and IAM.
- Access to Amazon Bedrock and necessary FMs.
- An operational Amazon Redshift instance populated with relevant data.
- Understanding of the Agentforce system and how to navigate it.
- Familiarity with Lambda function creation and deployment.
Final Steps
The integration involves several more steps, including setting up IAM roles, creating a knowledge base and Amazon Bedrock agent, configuring a Lambda function, and establishing a REST API via the API Gateway. Additionally, named credentials in Salesforce must be created to secure the connection and allow API communication underpinning the entire workflow.
As organizations continue to leverage multi-agent systems, such integrations will prove invaluable in automating workflows, enhancing efficiency, and delivering exceptional customer service.
Conclusion
This blog post illustrated how combining AI services on AWS with Salesforce’s Agentforce can revolutionize business processes. By utilizing Amazon Bedrock Agents for contextual understanding and employing Lambda functions and API Gateway to facilitate interactions, businesses can automate and optimize workflows like never before.
As AI capabilities continue to evolve, multi-agent systems will become more critical to enterprise automation strategies, making it essential to explore these possibilities further.
Stay tuned for our upcoming posts as we dive deeper into asynchronous integration patterns using Salesforce Event Relay and Amazon Bedrock.
About the Authors
Yogesh Dhimate is a Sr. Partner Solutions Architect at AWS, focusing on Salesforce partnerships.
Kranthi Pullagurla has extensive experience in application integration and cloud migrations.
Shitij Agarwal is a Partner Solutions Architect at AWS.
Ross Belmont and Sharda Rao are both Directors at Salesforce, driving platform innovation and market strategy.
Hunter Reh is an AI Architect at Salesforce, specializing in agent development.
Together, this team combines years of expertise to navigate the future of enterprise automation.