The Short Version
Gordon is Docker's AI agent built into Docker Desktop. It reads your Dockerfiles, compose files, and running containers, then suggests fixes, generates configurations, and executes commands with your approval. It's the most useful AI DevOps tool we've tested — if you use Docker, you should be using Gordon.
What It Does
Gordon operates at three levels:
-
Explain. Ask "why is my container crashing?" and Gordon reads your logs, inspects the container state, and explains the issue in plain English. It doesn't just paste stack traces — it tells you what's wrong and why.
-
Fix. Gordon proposes specific fixes: updated Dockerfiles, new compose configurations, environment variable changes. You review and approve before anything runs.
-
Act. With your permission, Gordon executes commands: rebuilding images, restarting containers, pulling new base images. Every action requires explicit approval.
What I Liked
-
Context-aware. Gordon reads your entire Docker environment — Dockerfiles, compose files, running containers, volumes, networks. It doesn't give generic advice; it gives advice specific to your setup.
-
Safe by default. Every destructive action requires approval. Gordon shows you exactly what it's going to do before it does it. No surprises.
-
Actually fixes things. I tested it on a container that was OOM-killing. Gordon identified the memory limit, suggested a specific increase based on the application's actual usage pattern, and offered to update the compose file. One approval and it was fixed.
-
Learns your patterns. Over time, Gordon learns from your Dockerfile patterns and compose configurations, making better suggestions for your specific setup.
What I Didn't Like
-
Docker Desktop only. Gordon lives inside Docker Desktop. If you manage containers via CLI on remote servers, you can't use Gordon there. It's a local tool, not a remote one.
-
Security concerns. Some security researchers have flagged that Gordon's broad access to your Docker environment could be a risk if the agent is compromised. Docker says all processing happens locally, but the concern is valid for highly sensitive environments.
-
Sometimes over-explains. When you just want a quick fix, Gordon's detailed explanations can feel slow. A "just fix it" mode would be welcome.
-
Limited to Docker. Doesn't help with Kubernetes, ECS, or other container orchestrators. If you're running K8s in production, Gordon only helps with your local development environment.
Who Should Use It
- Developers who use Docker Desktop: This should be on by default. It's free and it saves time on every debugging session.
- DevOps teams: Quick diagnosis of container issues without SSHing into servers or reading logs manually.
- Docker learners: Gordon explains what's happening in plain English. It's like having a tutor who knows your exact setup.
Who Should Skip It
- Kubernetes-native teams: Gordon doesn't help with K8s manifests or cluster management.
- Teams with strict security requirements: The agent's broad access may not pass security review in some organizations.
- Remote-only workflows: Gordon works locally in Docker Desktop, not on remote servers.
Bottom Line
Gordon is the most practical AI DevOps tool available. It doesn't try to replace your entire workflow — it makes the Docker part of it faster and less frustrating. If you use Docker Desktop, turn Gordon on. You'll wonder how you debugged containers without it.