Multi-Agent Development focuses on designing, building, and operating systems composed of multiple collaborating AI agents across two days. Attendees can take the course after Agentic Development Foundation as an intermediate specialization, or after Agentic Development Advanced for a deeper treatment of multi-agent architecture and operations. Day one starts with the foundations: weighing single-agent, workflow, and multi-agent tradeoffs across orchestrated, decentralized, hierarchical, and event-driven collaboration models. A lab reinforces the lesson by having attendees select and justify an architecture for a realistic use case.
From there, participants design role-based teams around supervisor, planner, worker, critic, reviewer, and router roles, assigning responsibility to avoid duplicated work and circular delegation. The day continues into communication and interoperability, covering message contracts, request/response and event-driven interactions, and capability discovery through MCP and API-based interfaces. It closes with a coordination module that applies orchestrator, supervisor, blackboard, market, and peer collaboration patterns to a shared task workflow.
Day two turns to production concerns. Attendees add reliability and consistency controls – idempotency, retries, fallbacks, and guardrails – to keep probabilistic agent behavior under deterministic control. They then instrument agent teams with tracing, dashboards, and human intervention points for approval, rollback, and escalation. A testing and evaluation module builds an evaluation harness using golden datasets, task rubrics, and LLM-as-judge scoring to catch regressions across model, prompt, and tool changes.
The course closes with scaling, governance, and production architecture, examining runtime choices from local development through Kubernetes and serverless deployments, cost and concurrency controls, and the governance practices enterprise teams need for auditability. Through hands-on labs, participants build progressively richer multi-agent workflows using modern protocols, frameworks, and testing methods.
Who Should Attend
Developers, AI/ML Engineers, Data Engineers, MLOps/DevOps professionals, architects, and technical leads
What Attendees Will Learn
Upon completing Multi-Agent Development, participants will be able to:
- Explain when multi-agent systems are useful and when simpler agent or workflow designs are better
- Design role-based agent teams with clear responsibilities, communication patterns, and control boundaries
- Implement multi-agent coordination using orchestrator, supervisor, handoff, and peer collaboration patterns
- Apply MCP, A2A-style interfaces, APIs, queues, and event streams to support agent interoperability
- Evaluate reliability, consistency, safety, and cost across distributed agent workflows
- Add observability, intervention points, testing, and recovery mechanisms to multi-agent systems
- Plan scalable multi-agent architectures for real-world enterprise and developer use cases
Prerequisites
Participants should have Python skills and experience with LLM applications. They should also be familiar with core agent concepts such as prompting, tool calling, structured outputs, context management, and basic agent workflow design.