Harness announced that it had acquired the Cosmos software factory, Auggie CLI, and Code Context Engine from Augment Code, along with the team that developed these products, in a deal whose value was not disclosed. The deal adds autonomous coding agents to Harness’s existing platform for managing software testing and deployment.
Harness will turn Cosmos into the Harness Cosmos Software Factory Agent, with plans to connect Augment’s understanding of source code to Harness’s delivery operations history. However, Cosmos is currently available, while capabilities for leveraging previous deployment failures or receiving feedback from downstream systems when validation fails have not yet been released.
From Requirement to Pull Request
Cosmos can turn a requirement or bug report into a pull request, as its agents write and test code inside isolated virtual machines. They can also handle failed tests and review comments on an existing pull request, then modify the code based on the results of previous attempts.
The system allows teams to clone agents and customize them to fit their workflows, while the Code Context Engine retrieves relevant context from the codebase during execution. According to Harness, developers remain responsible for approving and merging changes, with the ability to specify which tasks the agent executes automatically and which require human review.
What Is Supposed to Change in Practice?
Harness plans to connect the Code Context Engine to what it calls the Software Delivery Knowledge Graph, allowing Cosmos to receive information from deployment history and security scan results. If Harness agents discover a problem during testing, security checks, or deployment, they are expected to send it back to Cosmos along with the context needed to fix it.
Under this model, the agent’s role does not stop at creating code or opening a pull request; it could receive an actual deployment failure and make a new modification, instead of requiring the developer to diagnose the problem and create a separate task. This is connected to Harness’s research on what it called the AI speed paradox, in which accelerating code generation increases pressure on testing and release processes if downstream stages do not evolve at the same pace.
The Most Important Capability Has Not Arrived Yet
Harness has not specified an integration launch date, nor explained how context will be exchanged between the Code Context Engine and the software delivery knowledge graph. Therefore, the loop linking deployment failures to automated fixes remains a future plan, rather than a capability currently available to Cosmos users.
Permissions Remain an Open Question
Harness intends to subject Cosmos to the identity and policy controls in Agent Harness, but it has not yet clarified how permissions will be applied between the two systems. Cosmos currently operates inside isolated virtual machines with configurable review checkpoints, but the source does not specify which production data the agents will be able to access or which subsequent actions they will be allowed to perform.
certi.news analysis: The deal’s technical value lies not simply in adding an agent that writes code, but in attempting to close the loop between programming and operations. The success of this model will depend on the quality of the context reaching the agent, the limits of its permissions, and the clarity of human intervention points. Until Harness announces the integration timeline and permission-management mechanism, the deal’s most consequential component remains an engineering promise rather than a completed feature.