Agentic Engineering & ToolingKeynote20min
A Fleet of AI Agents, Each in Its Own Sandbox: Inside Docker's Agentic Platform
Docker’s Agentic Platform enables AI agents to autonomously read issues, code, test, and prepare pull requests in isolated sandboxes with secure identities. The talk covers orchestrated specialist agents, parallel workflows, and key production lessons on cost, hallucinations, human oversight, safety, and observability.
talk.summaryAiDisclaimer
Jean-Laurent de MorlhonDocker
talkDetail.whenAndWhere
Wednesday, October 7, 10:30-10:50
TBA 7
talks.roomOccupancytalks.noOccupancyInfo
What if your next pull request was opened by an AI agent, one that read the issue, wrote the code, ran the tests, and only pinged you when it was ready to merge?
That's how Docker's engineering team ships software today. Built on top of Docker Cloud Sandboxes, the Docker Agentic Platform gives every AI agent its own isolated environment: real shell access, a live Docker daemon, git, GitHub CLI, and a JWT-scoped identity. Orchestrators spawn specialists, architect, implementer, reviewer ; that work in parallel across sandboxed boundaries, coordinating asynchronously without ever touching each other's secrets or state.
We'll walk the hard lessons (context cost, hallucinations, knowing when to keep a human in the loop), and what it takes to make autonomous agents safe and observable in production.
That's how Docker's engineering team ships software today. Built on top of Docker Cloud Sandboxes, the Docker Agentic Platform gives every AI agent its own isolated environment: real shell access, a live Docker daemon, git, GitHub CLI, and a JWT-scoped identity. Orchestrators spawn specialists, architect, implementer, reviewer ; that work in parallel across sandboxed boundaries, coordinating asynchronously without ever touching each other's secrets or state.
We'll walk the hard lessons (context cost, hallucinations, knowing when to keep a human in the loop), and what it takes to make autonomous agents safe and observable in production.
Jean-Laurent de Morlhon
Jean-Laurent de Morlhon is SVP, GenAI Acceleration at Docker, where he focuses on making AI application and agent development simpler for developers.
Over more than nine years at Docker, he has held several engineering leadership roles, helping build and scale teams working on developer tools and infrastructure.
Today, his work focuses on GenAI developer experience, from running models locally to MCP, sandboxes and agentic workflows, with a particular interest in making emerging AI technologies practical, reliable and easy for developers to use.
Over more than nine years at Docker, he has held several engineering leadership roles, helping build and scale teams working on developer tools and infrastructure.
Today, his work focuses on GenAI developer experience, from running models locally to MCP, sandboxes and agentic workflows, with a particular interest in making emerging AI technologies practical, reliable and easy for developers to use.