From Solo Helpers to Team-Layer Coding Platforms
AI coding agents teams rely on are shifting from individual productivity tools to shared engineering infrastructure, where agents operate as coordinated services across projects, workflows, and approval gates instead of isolated assistants helping a single developer at a time. In the first week of June, three vendors pushed this transition forward together. Cognition released Devin Desktop, Microsoft introduced Rayfin at Build, and Augment Code announced Cosmos for every team plan. Each product targets a different layer of what now looks like a common stack: Devin Desktop provides a team console for AI agent management workflow inside the IDE, Rayfin governs which agent-built software is allowed into production, and Cosmos coordinates a fleet of agents across the lifecycle. Taken together, they signal that AI coding agents are becoming a standard part of team infrastructure, similar to how Git and CI moved beyond single-machine tools.
Devin Desktop: Turning the IDE into an Agent Command Center
Devin Desktop engineering focuses on pulling the coding agent into the center of day-to-day development work. Cognition describes it as the next generation of Windsurf, combining a full code editor with an Agent Command Center for managing local and cloud agents in one view. The same Devin agent can now run across desktop, cloud, and command-line surfaces, giving teams a consistent way to work with agents wherever code gets written or run. A new feature called Spaces groups agents by project so they can share context across sessions, pull requests, files, and tasks rather than starting from scratch each time. According to Cognition’s Theodor Marcu, “The question for engineering leaders is no longer whether to use AI — it is how to manage a growing fleet of agents working across their organisation simultaneously.” The answer here is a single environment that keeps existing Windsurf tools while adding shared coordination around them.
ACP and Agent Neutrality: Managing a Mixed Fleet from One Desktop
A key step toward team-layer coding platforms is managing many agents, not one. Devin Desktop does this through the Agent Client Protocol (ACP), an open standard that lets compatible third-party and in-house agents run alongside Devin inside the same environment. At launch, Devin Desktop can host agents such as Codex, Claude Agent, OpenCode, and other ACP-compatible tools on a shared dashboard. That means a lead engineer can assign one agent to refactor a legacy service, another to draft tests, and a third to triage issues, all visible within common Spaces and sharing context. This AI agent management workflow treats agents less like separate products and more like pluggable services in a single console. For teams that already mix vendors, the agent-neutral stance reduces the risk of lock-in and turns the IDE into a coordination plane rather than a window into one model.
Cosmos and Rayfin: Control Planes for Agent-Driven Delivery
While Devin Desktop lives in the developer’s seat, Augment Code’s Cosmos and Microsoft’s Rayfin sit higher in the stack, closer to CI/CD and governance. Cosmos is described as a control plane that spans triage, specification, implementation, review, testing, deployment, and feedback, coordinating specialist agents at each stage. Shared memory is central: what one agent learns during incident triage or testing does not disappear when its session ends, so the next agent continues with richer context instead of facing a cold start. Rayfin, announced at Build, focuses on what reaches production, acting like a gatekeeper for agent-built applications inside the enterprise. Together they mirror familiar patterns: Cosmos resembles a CI/CD control plane deciding which jobs run in which order, while Rayfin echoes policy and deployment layers that decide what is allowed to ship and under which rules.
Why Teams Now Expect AI Agents as Core Infrastructure
These launches show that AI coding agents teams use are crossing a line from optional helpers to expected infrastructure. Earlier competition centered on model quality and single-developer harnesses wrapped around them. Now, top systems cluster close enough on benchmarks that meaningful differentiation comes from how agents are coordinated, reviewed, and connected to production workflows. Team-layer coding platforms have to remember decisions across people, prevent agents from stepping on one another, and give humans clear points for judgment, much like reviewers on pull requests. Devin Desktop turns the IDE into a shared console, Cosmos acts as a fleet coordinator with memory, and Rayfin sets guardrails on what agents can deploy. For engineering leaders, the question is no longer whether to experiment with agents; it is how to make them safe, auditable, and dependable parts of the standard development stack.
