Cohere makes North generally available

Background

Agentic AI is a loose label for systems that do more than answer a question: the model plans a sequence of steps, calls other software tools, reads what comes back and keeps going until the task is finished. That is a larger ask than a chat window, and for big organisations the sticking point is usually less about raw capability than about control — where the data sits, which internal systems an agent is allowed to touch, and who can check afterwards what it did.

What happened

On 6 August 2025, Cohere made its agentic AI platform North widely available, moving it beyond the early deployments it had been running with a first set of customers ^[1]. Cohere framed the target as the "boring" multistep work inside companies: the routine, many-step processes that eat up staff time without needing much judgement ^[1]. The general availability came after a period of testing by a group of large enterprises ^[2].

Key details

The most distinctive part of the release is where North is allowed to live. Cohere says the platform is designed for secure deployments across on-premises infrastructure and hybrid clouds, and that it can run in a private cloud or on-premises ^[2]. It can be set up within a company's own environment ^[1].

That distinction matters. On-premises means the software runs on servers the customer owns and operates in its own data centres; a private cloud is a dedicated environment that is not shared with other customers; a hybrid setup splits workloads between the two. In each case the data stays inside a perimeter the organisation already governs, rather than sitting in a vendor's shared service. For a bank, an insurer or a hospital group, that is often the condition that has to be met before any agent is allowed near internal documents or records.

Cohere named companies using North as RBC, Dell, LG CNS, Ensemble Health Partners and Second Front ^[2]. Dell includes North in its Dell AI Factory stack ^[2], and North has been deployed within LG CNS in South Korea ^[2]. Together the list spans banking, hardware and IT services, healthcare revenue management and software — a spread of regulated and non-regulated buyers rather than a single design partner.

Why it matters

The commercial timing was tight. A week after the release, Cohere announced a $500 million funding round ^[1]. A generally available product with named enterprise customers is a different proposition to a pilot when a company is raising money and asking buyers to commit to a platform rather than a single model.

The wider point is about how agentic AI reaches large organisations at all. Most of the attention in this field goes to models that are consumed through an API or a chat interface. North's bet is that a meaningful share of enterprise work will be done by agents that sit inside the customer's own infrastructure, running multistep processes end to end, because the organisations with the most routine work to automate are also the ones with the strictest rules about where data can go. Whether that bet pays off depends on deployments like the ones Cohere listed: an agent platform is judged on whether it can be installed, governed and kept running inside a real company's systems, not on a demonstration.