Omar Marques | Lightrocket | Getty Images
Cohere co-founder Nick Frosst is surrounded by chatter about artificial general intelligence, or AGI. He’s perfectly happy to stay out of the conversation.
Founded in 2019, by ex-Google AI researchers, Cohere is valued in the billions of dollars and is one of the more high-profile names in the world of generative AI, which has exploded since OpenAI debuted ChatGPT in late 2022.
But it’s not a company that’s well known among consumers, who have swarmed to chatbots and other tools from OpenAI, Google and Perplexity. Rather, Cohere is all about business.
“I’m in meetings with companies in health care, banking and IT all the time,” Frosst told CNBC in an interview this week. “The questions I get are about securely automating tasks like HR, payrolls, research and fraud detection to drive productivity. No one has ever asked me about achieving AGI, let alone ASI.”
The latter is short for artificial superintelligence, or AI that significantly surpasses human intelligence. OpenAI and Anthropic have both made it their goal to achieve it.
In its latest funding round in July, Cohere raised $500 million at a $5.5 billion valuation, more than doubling its valuation from the prior year. Investors in the company include Nvidia, AMD, Salesforce and Oracle.
While that would historically be a huge price tag for a company that’s not even 5 years old, it’s a fraction of what investors are paying for OpenAI, valued at $157 billion in a round announced in October, and Anthropic, which CNBC confirmed this week is in talks to raise funding at a $60 billion valuation.
Some of Cohere’s chief competitors in the AI arms race offer products for both consumers and businesses. OpenAI, for instance, launched ChatGPT Enterprise in 2023, and Anthropic rolled out Claude Enterprise in September.
Frosst said Cohere’s preference for the enterprise is centered around the idea that large language models are best at automating tedious tasks and “being a co-worker.”
“Really, it’s an automation tool,” Frosst said. “When I think about my personal life, there’s actually not a ton that I want to automate. I don’t want to write text messages to my friends faster. I don’t want to respond to emails more efficiently in my own life. But in my work life, I really, really do want to do that.”
Frosst said, “I want to be free to think creatively and not be bogged down.”
Shortly after closing its funding round in July, Cohere cut about 20 jobs. A company representative said at the time that it was an “internal realignment” and that Cohere had a “clear vision for the future.”
That vision includes going all in on AI agents.
While the term AI agents isn’t neatly defined, it’s generally meant to describe AI services that go a step beyond chatbots. Agents are typically designed for specific business functions, rather than general purpose, and can be customized on the big AI models.
They can perform multistep, complex tasks on a user’s behalf and generate their own to-do lists, so that users don’t have to walk them through the process step-by-step.
Staying capital efficient
On Thursday, Cohere debuted its early access program for its AI agent platform called North, which is focused on allowing users with any level of technical background to “instantly customize and deploy AI agents” and do so “with just a few clicks,” the company said in a press release. Users can search for information across their organizations in multiple languages and in divisions with programs that weren’t previously connected.
That includes summarizing questions and answers in HR, speeding up the amount of time spent on finance reports, and automating some core business functions in customer support and IT.
Frosst said the platform can be used in any industry, but the company plans to target finance and health care, where data privacy and regulation are paramount.
Martin Kon, Cohere’s operating chief, told CNBC in March that by staying focused on enterprise AI, the company is able to run efficiently and keep expenses under control even amid a chip shortage, rising costs for Nvidia’s graphics processing units, or GPUs, and ever-changing licensing fees for AI models.
Frosst says those dynamics are still at play, allowing Cohere to be “more capital-efficient,” which is increasingly “of interest to investors.” Rivals with popular consumer-facing AI products, he said, use a lot of compute on “consumer applications and science projects.”
Although the sales cycle for enterprise AI can be longer, Frosst said, “the recurring business we’ve been able to create is something that’s really resonating with investors now.”
Competition is stiff and the technology is quickly evolving.
In October, Anthropic said its AI agents had the ability to use a computer like a human would in order to complete complex tasks. The feature, called Computer Use, allows its technology to interpret what’s on a computer screen, select buttons, enter text, navigate websites and execute tasks through any software and real-time internet browsing.
OpenAI reportedly plans to introduce a similar feature soon. And last year, executives from Microsoft, Meta and Google regularly touted their goals to push AI assistants to become increasingly productive.
Even without a consumer business, Cohere has to spend heavily on Nvidia’s costly GPUs, which are in huge demand for companies that are training models and running big workloads. In Cohere’s early days, the company secured a reserve of Google chips to help it pretrain its models. Over the past year, Cohere has moved more toward Nvidia’s H100 GPUs.
“We’ve increased our spend on them, because they’re working really well,” Frosst said.
