Case 04 / 06 — Selected Work
The agent app and the runtime beneath it, designed, researched, and validated in prototype.
Docker's bet on agents has two layers: cagent, the CLI multi-agent runtime that Gordon and agents run on (public as docker-agent), and dagent, the agent app where developers create, run, import and share agents with a marketplace (“Agent Fleet”). This case is the design-and-research narrative of the app layer that fronts the runtime.
The platform was researched while it was built. The Dagent UX Study ran ~11 moderated think-aloud sessions (Oct 2025) covering agent creation, running, and sharing, plus a continuous feedback instrument ("get the Agents from users" via URL or .yaml upload). Findings were synthesized against personas Priya (agent engineer), Marco (governance) and Sam (newcomer).
The runtime side was measured through cagent command events (~20.7M), and the prototype itself became an instrument: 369 agents created, 2,144 runs, 475 pulls across 96 users.
Sessions research forced the core IA decision: what is a session, what is a sandbox, and how do kits and marketplace listings relate. The answer shaped the New Session Composer direction: sessions as first-class, observable units (steps, artifacts, token budgets), with cost visible inline rather than in a billing page after the fact.
For the runtime, the concept was agents-as-composable-units: YAML definitions that are git-friendly and portable, with the app layer (dagent) providing creation, sharing and governance on top.
Each card defines the job, the design response, the objective, the measured outcome, and the trade-off made.
“Run multi-agent workflows from the terminal”
“Create, run and share agents without building from scratch”
“Bring an existing agent definition into Docker”
“Understand what people actually do with agents”
dagent numbers are prototype-phase by design; the runtime is open source as docker-agent at github.com/docker/docker-agent. Naming: dagent = app, cagent = runtime, Gordon = embedded assistant.
The honest framing matters: this is a design-and-research narrative of a platform in motion. Primitives moved because observed behavior said so, not because the roadmap said so.
Exported from the original Figma files with figx.
