From isolated prompts to ongoing work
Imagine an employee asking an AI agent to prepare a customer briefing. The agent searches internal documents, checks a database, drafts a summary, requests approval and updates a record in another system. The task may take minutes, or it may trigger a chain of actions that continues throughout the day.
That is a different product category from a chatbot that answers one question at a time. It also creates new operational risks. An agent can consume more computing resources than expected, call tools repeatedly or make decisions using context that is incomplete, outdated or scattered across different systems.
Cohere says in its announcement of North 2 that the new version of its enterprise AI platform is built around this change in how companies use artificial intelligence. The emphasis is not only on what a model can say, but also on how agents are managed, what they remember, where they run and which actions they are permitted to take.
North 2 introduces user quotas, rate limits and organization-wide spending caps. Its administration layer is designed to give businesses real-time visibility into usage, including token spending and configurable consumption tiers. Administrators can also set alerts before limits are reached.
For an enterprise, those controls could turn an unpredictable AI bill into something closer to a managed operating budget. A finance team might permit a research group to run sophisticated workflows while applying tighter limits to routine automations. An organization could also prevent a single user or agent from consuming resources intended for an entire department.
Memory as a workplace feature
The other central feature is memory. Cohere says North 2 allows agents to retain context across sessions, while shared libraries of knowledge and reusable skills can be applied across agents and automations.
The practical difference is significant. Without persistent context, employees must repeatedly explain the structure of a project, the meaning of internal terms or the preferences of a particular customer. With memory, an agent could begin a new interaction with an understanding of previous work, approved procedures and the information that a team has chosen to make available.
This does not mean that an agent automatically understands an organization. Memory is only useful when it is accurate, relevant and governed. A system that remembers an old policy or an incorrect customer detail can produce answers that sound confident while pointing people in the wrong direction.
North 2’s shared libraries suggest a more deliberate approach. Teams can collect knowledge and skills that are reusable rather than rebuilding the same instructions for every agent. In a large company, that could help standardize how agents summarize contracts, classify support requests or prepare internal reports.
It also raises questions that customers will need to answer. Who is allowed to add information to a shared library? How long should an agent retain a conversation? Can a person correct or remove a memory? Which parts of a customer record should be available to an agent handling a separate task? The value of memory will depend as much on those governance choices as on the underlying model.
Deployment where the data lives
Cohere says North 2 can run in the cloud, on premises or in fully air-gapped environments. The platform also supports Cohere models alongside models selected by customers.
That flexibility is aimed at organizations that cannot place all business information in a public cloud environment, including companies operating under strict security or regulatory requirements. An air-gapped deployment can keep an agent separated from external networks, while an on-premises option may give an enterprise more control over data processing and infrastructure.
Model choice is another part of the pitch. Businesses often hesitate to build workflows around a single model provider because changing models later can require costly redevelopment. A platform that allows customers to choose models could make it easier to adapt as capabilities, prices and corporate policies change.
Yet model portability does not automatically eliminate dependence on a platform. The surrounding agent harness, memory systems, permissions and workflow integrations can become just as important as the model itself. If a company builds its operating processes around North 2, the cost of moving away may still depend on how portable those surrounding components prove to be.
Guardrails for agents with authority
Cohere also describes security controls for prompts and responses. The platform can scan for personally identifiable information, detect prompt injection and restrict agents to authorized actions. Critical decisions can be routed to human users.
These features reflect a basic reality of enterprise automation: an agent becomes more useful as it gains access to more tools, but that access increases the consequences of mistakes. An agent that can read a document is one kind of system. An agent that can send a message, change a record or initiate a transaction requires a much clearer boundary between recommendation and action.
Human review can provide that boundary for higher-risk decisions. It can also make an automated workflow slower, which means companies will need to determine where oversight is essential and where routine actions can proceed automatically.
Cohere says North 2 holds SOC 2 Type 2, ISO 27001 and ISO 42001 certifications. Those credentials may help organizations evaluate the platform, but certification alone cannot show whether an agent will perform reliably in a particular company’s environment.
The cost of making agents dependable
North 2 arrives as enterprise AI moves toward persistent, multi-step systems. The challenge is no longer simply producing an impressive answer in a demonstration. It is keeping an agent within budget, giving it the right context, limiting its authority and making its behavior visible to the people responsible for the business.
Cohere has not provided performance benchmarks or public pricing for the release. It says costs depend on factors including deployment scale, infrastructure, usage, support and customization. That leaves customers with an important evaluation task. They will need to measure not just model quality, but also whether North 2 reduces duplicated work, prevents waste and improves the reliability of real workflows.
The most consequential AI systems in offices may therefore be the least theatrical. They will quietly remember a team’s processes, operate within a defined budget and ask for help when a decision exceeds their authority. North 2’s larger proposition is that enterprise AI will be won through this layer of operational discipline, where useful agents become manageable parts of everyday work.
Maya Lindqvist is not a person. No notebook, no deadlines, no face behind the name — just a byline this newsroom publishes under. Here is the production line underneath it, because a name beside a portrait reads like a journalist, and this one is not one.
The models. Writing: gpt-5.6-luna and qwen3-max. Out on the live web: gpt-5.6-luna and gpt-5.6-terra. Pictures: gpt-image-1 and gpt-image-1-mini. Swap one in the newsroom and this line swaps with it — it is read off the machines, not typed here.
How a story is made
- Research. The searching model reads around the story, pointed at primary sources — the filing, the post, the repository — rather than at somebody else's write-up of them.
- Writing. The writing model drafts it against what was found, at Maya Lindqvist's usual length and in Maya Lindqvist's usual register.
- The loop. A reviewer reads the draft and sends it back with notes. Then reads it again. A piece can go round several times before it leaves the building.
- Enrichment. A quotation has to appear word for word on the page it is taken from. A chart may only use figures that appear in the source it cites. Whatever fails is dropped, and the reason is kept.
- Fact check. A last pass hunts for claims the article makes and its sources do not.
- A human stop. Sensitive subjects are held for a person to read before publication, and a person can kill any of it at any point.
If that sounds less like a newsroom and more like a factory: quite. It is called Press Factory.
This article was generated using AI and published automatically without human pre-publication review.
How this article was made
The article was produced by the Grandmonts Media News Engine using automated research, drafting and verification workflows. No human editor reviewed the article before publication. Grandmonts Media remains responsible for the published content. Errors can be reported at office@grandmonts.cz.