Docker is hosting a workshop on cagent, its tool for building multi-agent AI systems using declarative YAML files. The session targets developers and DevOps engineers exploring how to orchestrate AI agents like microservices.
Launched in September 2025 and included in Docker Desktop 4.49+, cagent lets teams define agents, models, and tools in configuration files, then run them locally via Docker Model Runner or against cloud LLMs like OpenAI and Gemini. Teams can compose multi-agent systems with delegation hierarchies and push them as OCI artifacts to Docker Hub.
The workshop covers agent definition and configuration, multi-agent team orchestration for complex tasks, deployment using Docker workflows, and tool integration via Model Context Protocol (MCP). MCP handles tool negotiation between agents, with existing integrations for GitHub, search, and other services.
Docker positions cagent as "Docker Compose for AI agents," applying its Build, Ship, Run philosophy to agent systems. The approach mirrors how containers transformed application deployment: composable units that can be defined once and deployed anywhere Docker runs.
The trade-offs matter here. Giving agents system access via tools like shell and filesystem creates risks when LLMs make poor judgment calls. Docker's framework includes guardrails, but teams need to think carefully about what capabilities they expose.
The workshop is part of Docker's broader AI push. Related sessions include an MCP workshop at MCPconference London on February 12 (co-located with ContainerDays), a webinar on MCP for multi-agent systems on February 17, and a secure images session on February 19.
Registration is open via Google Form. Docker says it's keeping the session small for hands-on support. Requirements: laptop, Docker installation, no prior agent experience needed.
What this means in practice: If you're evaluating agent frameworks, cagent offers a familiar deployment model. The Docker ecosystem gives it distribution advantages. The real test comes when teams try to build production agent systems and hit the complexity of context flow and tool integration. We'll see if the container analogy holds.