Augment Code launches Cosmos, an operating system for multi-agent software development workflows
Augment Code has released Cosmos into public preview, positioning it as an operating system for agentic software development. The platform coordinates AI agents across the full software development lifecycle with shared memory, multi-model routing via their Prism system that claims 20-30% token savings, and what the company calls specialized agents that learn from team feedback.
Augment Code launches Cosmos, an operating system for multi-agent software development workflows
Augment Code has released Cosmos into public preview, positioning it as an operating system for agentic software development. The platform is designed to coordinate multiple AI agents across the entire software development lifecycle rather than serving as a single coding assistant.
Architecture and capabilities
Cosmos runs agents either in customer environments or Augment's cloud infrastructure. The system includes shared context and memory across agents, connections to development tools, and multi-model routing through what Augment calls Prism. According to the company, Prism delivers 20-30% token savings without quality degradation by selecting appropriate models for different tasks.
The platform introduces what Augment describes as "specialized agents" that store information from developer feedback. The company cites their internal testing agent "Milo" as an example, claiming it improves through conversational coaching rather than upfront context loading.
Workflow structure
Cosmos reduces the software development process to three human checkpoints, according to Augment:
- Prioritization review: Agents monitor feedback channels and propose daily priorities for human approval
- Spec and intent review: Humans review specifications before agents proceed with implementation
- Contextual understanding: A review experience focused on surfacing assumption changes rather than line-by-line code inspection
The company introduces what it calls "deep code review" — an agent-driven review process optimized for recall rather than precision, designed to catch all potential bugs when the reviewer is an AI agent rather than a human.
Availability and positioning
Cosmos is available now in public preview for MAX plan users. Pricing for the MAX plan was not disclosed. VP of Engineering Vinay Perneti wrote that the company is releasing early with "rough edges" to learn with teams experiencing what he describes as a disconnect: widespread individual agent adoption without corresponding organizational productivity gains.
The announcement frames Cosmos as infrastructure for what Augment calls "small teams of people working with large teams of agents." The company references Claude Opus 4.5's November release as a turning point when "a large portion of serious software engineers agreed it no longer makes sense to write most code by hand," though this claim is not independently verified.
What this means
Augment is betting that AI coding tools need orchestration layers rather than more powerful individual agents. The company's emphasis on model-agnostic routing addresses a real concern as frontier models become increasingly expensive for routine tasks. However, the core premise — that agents with memory and feedback loops solve organizational adoption challenges — remains unproven at scale. The public preview will test whether coordinating multiple specialized agents delivers measurable team-level productivity gains beyond individual developer efficiency.
Related Articles
Augment Code launches Prism router: 20-30% cost reduction routing between Claude Opus 4.7, GPT 5.5, and cheaper models
Augment Code released Prism, a model routing system that selects between frontier models and cheaper alternatives per user turn. On internal benchmarks, Prism matches Claude Opus 4.7 and GPT 5.5 quality while reducing per-task costs by 20-30%, translating to approximately $20,000 monthly savings for teams sending 10,000 requests.
OpenAI adds Tamagotchi-style pets to Codex Mac app with custom creation feature
OpenAI added a /pet feature to its Codex Mac app that displays animated companion creatures in the interface. Users can choose from preset options or create custom pets that provide status updates while Codex runs in the background.
Perplexity's Mac-Native 'Personal Computer' Platform Claims $2.8B in Labor-Equivalent Work
Perplexity CEO Aravind Srinivas revealed that the company's Mac-native Personal Computer platform has performed more than $2.8B in labor-equivalent work for Pro, Max, and Enterprise subscribers since launch. The announcement follows Apple CFO Kevan Parekh citing Perplexity as an example of developers building enterprise-grade AI assistants on Mac during Apple's Q2 2026 earnings call.
Meta building personal and business AI agents on top of Muse Spark model
Meta is developing AI agents for personal and business use that will run continuously to help users achieve goals, CEO Mark Zuckerberg said during the company's Q1 2026 earnings call. The agents will build on Meta's newly-released Muse Spark model from Meta Superintelligence Labs.
Comments
Loading...