Augment Code launches team-based AI development platform
Augment Code Computing Inc. announced Cosmos on Thursday, a service designed to coordinate AI agents across engineering teams rather than deploying them to individual developers in isolation.
The platform addresses a practical problem: while individual developers using AI agents see productivity gains, teams as a whole experience uneven results. One engineer might offload tedious work to an agent and ship faster. Another can't, because the context of what the first engineer accomplished isn't visible to their agents.
Vinay Perneti, Vice President of Engineering at Augment Code, described the progression: "2024 was mostly chat, 2025 is agents, but 2026 is going to be agents for teams."
How Cosmos works
Cosmos operates as a software delivery platform where humans and agents collaborate with clear boundaries. Agents handle execution and autonomous work. Humans retain control over prioritization, specifications, intent validation, and high-risk decisions.
A developer writes a specification as they would for a junior coder - describing what they want, key elements, and the primary goal. A validator agent reviews the spec against the codebase and company practices, then clarifies anything that needs updating. This keeps agents aligned and prevents them from drifting into poor long-term decisions.
Cosmos learns across every interaction. When a Slack message corrects an agent's approach, the system maintains a memory of that correction. Patterns, best practices, and lessons from one session carry forward to the next, shared across all agents and team members.
Cross-team access without bottlenecks
Cosmos extends beyond the engineering team. A data team can configure a read-only analytics expert with proper credentials. The sales team then accesses that expert directly through Cosmos to query business intelligence - no need to request data from the analytics team or manage separate tools.
The data team retains full control. Updates they make propagate immediately to everyone using that expert. The system stays governed and current without creating approval bottlenecks.
An adviser function helps users identify the right expert for their task. In organizations with dozens of agents and workflows, users describe what they need, and the adviser routes the work to the correct agent automatically.
Perneti said engineering leaders he spoke with adopted agents top-down across their teams, but remained unsatisfied with overall team throughput. Cosmos targets that gap - the difference between individual productivity gains and actual team acceleration.
Your membership also unlocks:
