Cohere has unveiled North 2, an update to its enterprise-focused agent harness that aims to make AI agents safer and more consistent for business use. The platform introduces skills—documented standard operating procedures that capture the best steps for completing tasks—along with automations, which are drag-and-drop templates for customizing workflows. A new libraries feature lets teams share knowledge assets so agents work from the same datasets, while memory maintains context across sessions.
Security and governance are the headline additions. North 2 lets administrators assign access control lists to libraries and skills, preventing lower-privilege users from pulling proprietary data simply by asking. Flow control functionality gives visibility into which departments are consuming the most tokens, helping enterprises manage spend. The harness can be deployed on-premises in an air-gapped environment or in the cloud, and Cohere says it works with whatever models enterprises prefer.
The update positions Cohere against a crowded field of enterprise agent platforms from OpenAI, Meta, and others. While those vendors have emphasized autonomous capabilities, Cohere is leaning into guardrails and accountability—offering a more locked-down alternative that still resembles familiar tools like Claude Code or OpenClaw, but with stricter access controls built in. The Register notes the competition has intensified in recent weeks as major cloud providers push their own enterprise agent offerings. [1]