Enterprise agent platforms usually make you pick two of three: smart enough for real work, secure enough for legal, cheap enough at scale. Cohere’s argument with North 2, announced on October 5, 2026, is that the platform layer can close that gap instead of forcing each company to engineer around it. The launch post describes a full-stack agentic platform shaped by what the company says were production deployments in finance, healthcare, telecom, manufacturing, energy, and the public sector.
The harness is the headline
The most consequential piece for builders is a redesigned agent orchestration system meant to handle multi-step processes on its own, with humans kept in the loop where it matters. Around it, Cohere has assembled the pieces most teams end up rebuilding in-house: Skills (reusable capabilities agents can call), Libraries (shared org-wide assets), Memory that persists across sessions, and Automations with a drag-and-drop workflow builder and real-time monitoring.
That reusable-capability pattern is worth pausing on. If Skills work as described, the unit of sharing inside a company stops being the prompt and becomes the capability — closer to how internal tooling platforms treat a well-scoped API. New connectors ship for Slack, SharePoint, OneDrive, Outlook, Exchange, Jira, Linear, Notion, and GitHub, with financial data providers like PitchBook, FactSet, and S&P Global planned.
Deployment location is still the enterprise wedge
North 2 can be self-hosted, run in a VPC, hybrid, or on-premises — including fully air-gapped. That’s the same bet we looked at in Sovereign AI, One Year In: for regulated buyers, where the model runs is a product feature, not an infrastructure footnote. Cohere backs this with SOC 2 Type 2, ISO 27001, and ISO 42001 certifications, plus guardrails, per-agent autonomy policies, red-teaming, and change logs for observability.
Bell Cyber and LG CNS are the cited customers. Jawid — rather, Jawed Ahmad, Bell Cyber’s Chief Technology and AI Officer, says in the post that North lets them bring agentic AI into security operations with sovereignty requirements built in from the start. Treat those as partner quotes from Cohere’s announcement, not independent evaluations.
Governance gets real controls
North Admin is the part that will decide whether this survives procurement. Admins can set roles and permissions, control which models are available to whom, integrate existing IAM systems, and define consumption tiers for token spend with alert thresholds — at the user and agent level, not just org-wide. That granularity matters because agent cost is a distribution problem: one team’s over-eager agent can blow the budget for everyone.
On the efficiency side, Cohere claims that running its models on NVIDIA Blackwell and Hopper hardware yields more tokens per second per node, and that North “generates more insights, for fewer tokens.” The tokenomics claim comes from a vendor collaboration, so verify against your own workload before budgeting around it.
What builders should take from it
Model-agnostic is the other notable commitment: bring your own models or use Cohere’s. Combined with deployment flexibility, the pitch is that the platform value sits in the harness and admin layer, not in locking you to a model.
The grounded takeaway: North 2 is worth a close look if your roadmap includes agents that touch internal systems in a regulated environment — its per-agent autonomy policies and spend controls are the features to probe in a demo. The supplied announcement doesn’t include pricing, benchmark data, or independent security audits, so those are the gaps to fill before committing.