A Virtual Agent Team at Docker: How the Coding Agent Sandboxes Team Uses a Fleet of Agents to Ship Faster

Why It Matters

The Fleet turns repetitive DevOps work into self‑servicing AI processes, accelerating release velocity and improving software reliability while freeing engineers to focus on high‑value decisions.

Key Takeaways

• Docker's Fleet runs seven autonomous AI roles in CI pipelines
• Skills are persona‑based markdown files, not static scripts
• Local‑first development cuts debugging time from minutes to seconds
• Automated agents generate daily release notes and weekly tech‑debt PRs
• Ralph‑loop separates code generation from AI‑driven review for quality

Pulse Analysis

Docker’s Coding Agent Sandboxes provide microVM isolation that lets AI agents operate safely on developers’ machines and CI runners. The new "Fleet" composes seven distinct roles—build engineer, project manager, product owner, CLI tester, performance tester, upgrade tester, and software engineer—each described by a skill file that outlines a persona, responsibilities and permissible tools. By treating skills as roles rather than static scripts, Docker enables agents to exercise judgment, load foundational knowledge, and collaborate much like a human team would, all while staying within the secure sandbox environment.

The practical impact of this design is speed and consistency. Engineers iterate on skill files locally, seeing instant feedback, before the same definitions are deployed to CI pipelines. This local‑first approach eliminates the costly commit‑push‑wait cycle typical of traditional CI automation. The Fleet runs nightly across macOS, Linux and Windows, automatically testing the CLI, detecting regressions, and filing deduplicated issues. It also produces daily release‑note summaries for Slack and weekly tech‑debt pull requests, dramatically reducing manual toil and keeping the codebase clean. The Ralph‑loop further refines quality by separating code generation from AI‑driven review, mirroring classic QA principles.

Beyond Docker, the Fleet showcases how AI‑driven agents can be orchestrated as a virtual development team. By encapsulating expertise in reusable skill files and leveraging a composable loop for iteration, organizations can scale repetitive DevOps tasks without sacrificing oversight. The model promises faster feedback loops, lower operational risk, and a path toward more autonomous CI/CD pipelines, while still preserving human control over merges and strategic decisions. As AI agents mature, similar role‑based frameworks could become a standard component of modern software delivery stacks.