Docker, Inc

Most AI evaluation discussions focus on benchmarks and scoring. The part that quietly breaks reproducibility is the execution environment. Docker Captain Karan Verma explored whether Docker Sandboxes and SBX Kits could make evaluation workflows easier to rerun, inspect, and compare. That question evolved into building the SBX AI Evaluation Kit, an open-source Mixin Kit focused on repeatable execution and structured evaluation records. Read more:

Docker CTO Tushar Jain will speak at WeAreDevelopers North America, co-hosted by Docker. In this main stage talk, Tushar will talk about how to govern the runtime, not the agent. He will build the case for a runtime below all your agents: the layer below all your harnesses, covering agent execution, tool calls, credential usage, and scoped permissions. This session will show you how to get control and visibility into all your agents, no matter the underlying model or harness. September 23-25 in San Jose. Build what's /next. https://bit.ly/4gAFLqe

Most developers running coding agents in YOLO mode have no idea what the agent can actually reach. Product Manager Eric Jia and Staff Software Engineer Kevin Wittek built Docker Sandboxes. In this demo, they walk through exactly how to use them to set up the boundary your agents actually need. The demo covers credential injection, kits, how sandboxes are different from containers, and a brand new SSH feature that lets you use Claude Desktop and VS Code with full sandbox isolation underneath. Watch the full video → https://lnkd.in/eXKXuU3N

If you want to move past agentic pilots, you need answers for the security review. Where did the agent's base image come from? Which tools did it call? What did the approval chain look like? Most rollouts stall when there aren't good answers for these questions. Bjorn Hovd, CISSP takes the stage at WeAreDevelopers World Congress North America and covers what it takes to close that gap in his lightning talk: hardened base images, vetted MCP servers, sandboxed sessions, and an audit log for every action. Trust travels with the agent at every step. Come see how this enables developers to move fast with agents, safely. September 23-25 in San Jose, co-hosted by Docker. Find us at Booth # 101. Build what's /next. Tickets: https://lnkd.in/duyFhgtm

Mark Cavage will be giving a keynote at WeAreDevelopers World Congress North America in San Jose, co-hosted by Docker. Mark’s talk, “Manufacturing Trust: Speed and Safety in the Age of Agents,” is about the shift we’re all experiencing: agents are actors, and trust that used to live with people and process now has to live in the system itself. That means isolating where agents run, limiting what they can reach, recording what they do, and making failures reproducible. Get that right, and you can finally run agents the way you actually want. Keynote is September 24th. Conference runs the 23rd through the 25th. Come find Mark after his keynote to talk about what’s next.

80% fewer high-severity vulnerabilities and zero critical CVEs. That's what happened when IDOM, one of Japan's largest vehicle retailers, standardized on Docker Hardened Images (DHI). Their approach was straightforward: substitute standard images with the DHI catalog and ensure that every dev team was using the same secure foundation. Read the full case study here:

You'll run more than one model. You'll run more than one harness. But you should only have to write a policy once. Srini Sekaran makes the case of having a single runtime layer below all your agents. Tushar Jain, Docker CTO, talks about this further at WeAreDevelopers World Congress North America (co-hosted by Docker). Sept 23–25, San Jose. Read more:

Agent adoption is “kind of the wild west right now”. Everyone is using agents on their laptops in their own way, with access to company files and every connected system. No central oversight, no consistent controls, no visibility into what any of it can reach. Zach Lloyd (Warp) on what it takes to move past the free-for-all phase, and what the rest of the CISO panel says organizations need to put in place before it becomes a much bigger problem. Watch the full panel episode where Zach joins Mark Lechner (Docker), Gavriel Cohen (NanoClaw), and Moriah Hara (CISO Next Gen) → https://lnkd.in/gjydtD8Y

The Agent Hall of Shame winners have been selected. Congratulations to our winners! We asked for your most unhinged coding agent experiences. You gave us so many great stories, including infrastructure chaos, invented conspiracies, and an agent that declared war on itself. Want to swap agent horror stories in person? We’ll all be at WeAreDevelopers World Congress North America, September 23-25 in San Jose, co-hosted by Docker. Tickets here: https://bit.ly/4e7b7nD

Another exciting addition to WeAreDevelopers North America, co-hosted by Docker! Michael Yuan and Vivian Hu of Second State walk through an open-source agentic device, and a voice-first AI-native controller for agents like Claude Code and Codex. They cover the full stack from PCB design and ESP32 firmware to the WebSocket architecture that bridges device to agent runtime. Add this to your list if you want to understand where the physical layer of agentic computing is going. September 23-25 in San Jose. Build what's /next. Session details: https://bit.ly/4xvqRr8