Cohere unveils latest AI platform with more agentic features for businesses

Cohere Inc. is releasing a new version of its North artificial-intelligence platform in a bid to win more corporate and government customers in the fiercely competitive market for enterprise AI applications.

The Toronto-based company, which builds large language models, released the first version of North last year. Cohere chief AI officer Joëlle Pineau said the initial version functioned largely as a chatbot that could pull internal documents and handle some automated workflows.

“It was essentially a generative model,” she said. “Today, we’ve actually put in a lot of the features that allow agentic behaviour.”

Customers can use the latest version, called North 2, to build agents to carry out more complex multistep actions, such as resolving customer issues, conducting financial analysis and coding. Ms. Pineau said some features of North 2 have already been shipped to customers.

Bell, Cohere release AI model designed to aid cybersecurity investigations

North acts as a kind of software layer around AI models, and at its core is what’s called an agent harness, an increasingly common term in the industry. The harness manages agents and connects with other applications to complete tasks.

“The model is like the engine, and the harness is like the vehicle,” Ms. Pineau said. “If you have a motor, you’ve got a ton of power. But without a vehicle, you’re not going anywhere.”

Cohere has scored Bell Canada, LG CNS Co. Ltd. and Dell Inc. as North customers, but the company declined to identify how many clients it has in total. “We have dozens of companies, and some of them are up to tens of thousands of users,” Ms. Pineau said.

Founded in 2019, Cohere focuses on businesses, regulated industries and public-sector clients for whom privacy, security and control are crucial. Rivals OpenAI and Anthropic are gunning for these clients, too.

Cohere, Open Text to pitch Canadian-made AI agent tools to enterprises

OpenAI, for one, offers ChatGPT Enterprise for businesses and is rolling out a new agent product called Dots, which the company says can monitor customer feedback and suggest improvements, fix software bugs and update documents, among other tasks. Microsoft Corp. is selling its 365 Copilot product suite, while French AI company Mistral is making inroads with Canadian businesses.

Cohere is much smaller than its U.S. competitors, too, though it is bulking up through a merger with German AI company Aleph Alpha, giving it more access to the European market.

AI agents have dominated headlines in recent months for misbehaving and attempting to hack into organizations during evaluations, and sometimes succeeding. In the corporate world, companies want assurance that agents will carry out work correctly, access the right documents and tools, and not corrupt anything in the process.

Cohere said multiple security features are built into North. Agents can be designed to take only authorized actions and seek human input for more critical tasks. Customers can set parameters around the information agents can access and which employees can deploy certain agents.

Cohere, Aleph Alpha combine to target enterprise AI market

Companies can also monitor costs and set usage limits, reflecting concerns about rising AI expenses. The issue is compounded by AI agents, which can spend a lot of time trying to complete a task – and not necessarily in the most efficient way – pushing up usage costs.

North is model-agnostic, meaning customers can use it with any AI model, even those from Cohere’s competitors. “We actually derive a sizable portion of our revenue from the harness,” Ms. Pineau said.

Training AI models is an expensive endeavour, with OpenAI, Anthropic and Google spending billions and racing to develop more sophisticated LLMs. Ms. Pineau said Cohere sees value in continuing to train models and build the agent harness around them. “The hypothesis is you do a much better version of both when you develop both,” she said.