Free iQ MuleSoft Study Group (10/23): Evolution of Agent Building

Sravan Nerella returns to the Integration Quest MuleSoft Study Group for Evolution of Agent Building, exploring how developers can approach AI agent development and how the trade-offs change between custom implementations, agent frameworks, and managed platforms. Sravan begins with a key architectural question: how much control do you really need? He contrasts deterministic API development with the non-deterministic behavior of AI agents and breaks down core concepts including LLMs, tools, memory, agent loops, harnesses, and sub-agents. The session looks at why building agents directly in MuleSoft can become complex when developers take ownership of the loop, testing, debugging, observability, retries, permissions, and edge cases. Sravan then introduces his Agent Composer custom connector as a way to reduce that boilerplate while keeping agent development inside MuleSoft. The live demo walks through configuring an LLM provider, agent instructions, MCP servers, tool filtering, permissions, retries, memory, and A2A details. Sravan also demonstrates streaming responses and tracing, showing the LLM plan, tool calls, requests, responses, and execution flow behind the agent. From there, Sravan moves into agent management and compares Agent Fabric v1 with Agent Fabric v2. He explains how Agent Script and guided determinism expand what developers can do in v2 while also discussing remaining considerations around sub-agents, sequential execution, streaming, and visibility into external agent runtimes. Whether you’re a MuleSoft developer, architect, consultant, or exploring agentic development, this session offers a practical look at the evolution of agent building and the trade-offs between control, development speed, and management. Agent Composer GitHub: https://github.com/AskNerella/agent-c... Connect with Sravan Nerella: LinkedIn:   / sravan-nerella   Medium:   / sravannerella007   YouTube:    / @sravan-nerella   MuleSoft Community Slack: https://mulesoft-community.slack.com