Overview

This guided hands-on workshop walks attendees through the key stages of building enterprise-grade AI systems—starting with foundational LLMs and progressing toward enterprise-aware, interoperable, and orchestrated multi-agent architectures.

In enterprises, multi-agent does not automatically mean that agents are wired together or coordinated by a broker. Multiple agents can exist independently, serve different purposes, and never interact—and this is still a valid, and often desirable, multi-agent design. Orchestration is a later-stage choice, introduced only when agents must collaborate to solve a shared problem or workflow.

In this workshop, we explore multiple architectural patterns for building realistic multi-agent strategies.

For simplicity, the workshop is modular, and each stage in the multi-agent architecture can be leveraged independently based on business requirements. The workshop is designed to provide a practical, end-to-end mental model for transforming business processes with AI.

Maturity Journey Diagram

Time may not allow you to explore all workshop modules today. Each module is self-contained, so you can choose any one to explore on its own.

We recommend these modules to start with:

  • Stage 0: This is a fun exercise and demystifies AI
  • Stage 1: This is an easy and insightful exercise with Mule AI
  • Stage 4: The Document AI module (Salesforce Data360 with Agentforce)

Stage 0: Foundational Gen AI App:

Build and deploy a foundational Gen AI app that talks directly to a raw LLM, with no grounding, memory, or enterprise controls. This stage clarifies what an LLM can do on its own—and, just as importantly, what it cannot. You will develop a clean mental model of where “pure GenAI” ends and where real applications & platforms must begin.

Stage 1: From LLM call into an enterprise AI capability with Mule Inference Connector

Build and deploy a real, production-style LLM inference endpoint in Mule, exposed as a stable HTTP API and running on CloudHub. You will see how AI execution gains security, isolation, observability, and managed scaling the moment it is centralized behind the platform. You leave with the steps for creating reusable enterprise inference service that any application—or future agent—can safely consume without rework.

Stage 2: From Inference to Enterprise-Aware AI with Mule & MCP

Extend the Mule-hosted inference capability so the system can consult authoritative enterprise data at runtime instead of guessing. You build an MCP server as a live source of truth and connect it to the LLM, making responses fact-grounded, explainable, and controllable. You leave with a true agent foundation—where intelligence is separated from knowledge and access is governed by the platform, not prompts.

Stage 3: From Enterprise-Aware AI to Discoverable AI Agents (A2A + MCP)

Wrap your enterprise-aware AI capability in the Agent-to-Agent (A2A) protocol, turning it into a standardized, discoverable service that other machines—not just humans—can call. You deploy a real A2A server on CloudHub that exposes skills via an Agent Card while hiding internal reasoning, tools, and data access. You leave with a clear mental model of how AI capabilities are exposed machine-to-machine, how discovery and interoperability work without tight coupling.

Stage 4: Data Driven AI Agents with Data360 & Agentforce360

Leverage Data360 to power Salesforce Agentforce within real employee and customer workflows. You see how enterprise agents operate with data, identity, and guardrails in place—and leave with a repeatable blueprint for embedding AI agents into Salesforce business processes.

Stage 5: Orchestrating Multi-Agent Systems with Mule Agent Fabric Broker

Explore Mule Agent Fabric Broker to centrally discover, govern, and orchestrate multiple specialized agents across the enterprise. You explore how routing, coordination, and policy enforcement are defined declaratively—without hardcoding agent-to-agent links. You leave with a clear model for scaling from single agents to governed multi-agent systems without creating agent sprawl.

Workshop Environments

1. Salesforce

2. Heroku:

3. GitHub:

4. Groq LLM:

5. MuleSoft:

The instructions for Mule Remote Desktop will be sent to you via mail. However following is a summary:

  1. Install the lightweight Windows App on your laptop: https://modules-library.workshops.mulesoft.com/setups/setup/remoteDesktop.html
  2. Click the link in your mail, Restart Remote Desktop Instance
  3. A window will pop up - as your Remote Desktop instance is restarted - after a few minutes you will see Public and Private IP address displayed in your pop up.
  4. Copy the Public IP and use it to connect (OS specific instruction in #1 above. Your instructor will provide you with username/password).

Thank You