Reviewers see Docker as a dependable way to make software run the same in development and production, cutting down on “works on my machine” problems and letting teams isolate services with different dependencies. They also like using it to safely try software without installing it directly on their machines. The main complaints are practical: image size can grow too easily without clear warnings, build cache behavior can be opaque and costly when small changes invalidate layers, and mounted-volume performance on macOS still feels slow on larger projects.
Docker is the reason "well, it works on my machine!" stopped being a viable excuse. Containers gave every developer the same box to build in, and that box became the default unit of shippable software.
Today Docker is recognizing that the shape of the next default needs to be different.
Agents aren't workloads, they're actors. They read your files, call your APIs, and make decisions you didn't script.
A container isolates a process; it doesn't govern a thing with intent.
So the new Docker is built around trust:
A trusted supply chain for agents to build from
Docker Sandboxes to run Claude Code, Codex and other agents in MicroVMs, locally or in the cloud, with deny-by-default networking
Docker AI Governance to set sandbox, network and MCP policy once and enforce it on every machine
...plus a new brand and a redesigned Moby.
So if you're running coding agents today, what's your isolation story? A VM, a container, a prayer? What preconditions do you require to let an agent run unattended? 😬