TORONTO — Cohere plans to launch a new suite of workplace AI assistants and customize smaller models for clients, as the startup looks to capture market share against larger competitors.

In a letter sent to staff and shareholders, CEO Aiden Gomez said the Toronto-based firm is developing a new line of products that clients can install more easily. 

The firm will offer the tools to launch AI assistants that workers can use to search for information, write documents and answer questions, Gomez said. The automated helpers will plug into clients’ existing data and business systems like customer relationship management tools. Eventually, Cohere plans to offer agents, which can go out and do tasks for their users, like booking appointments or resolving customer service complaints.

A number of technology companies already sell agents, based on large language models (LLMs) from companies like OpenAI, Anthropic and Cohere. But some generative AI startups have themselves launched such assistants, as they look for more ways to generate revenue. 

At The Logic Summit in October, Cohere co-founder Nick Frosst signalled the firm was prepared to go beyond its core models. “We’ve created LLMs,” he said. “We’ve also created a variety of tools for making use of those LLMs.” When customers face challenges building atop the generative tools, “you can definitely expect us to try to make that easier for them.”

In his letter, Gomez also claimed Cohere is avoiding a growing challenge its competitors are facing with improving their models. OpenAI, Anthropic and Google have reportedly seen the rate at which its foundation models get better slow down, as developers hit the limits of gains provided by adding more data and processing power. 

Instead of “simply buying more supercompute,” Cohere has instead focused on training with better-quality data, and synthetically-generated information, Gomez said. “Our strategy has allowed us to be an order of magnitude more capital efficient than our competition.” The startup has raised US$940 million to date, according to PitchBook data, compared to OpenAI’s US$29.9 billion and Anthropic’s US$12.8 billion.  

Cohere has zeroed in on selling its models for businesses, using their most “sensitive and valuable data assets,” Gomez said. He touted Cohere’s ability to differentiate itself on trust as a “technology partner” rather than supplier.

An “off-the-shelf consumer AI chatbot” isn’t enough, Gomez wrote in the letter, adding that customization is highly sought after by corporations. 

The AI company has already bolstered access to its products through partnerships with cloud providers like Amazon Web Services, Microsoft, Google and Oracle. It also works with consulting giants McKinsey and Accenture. 

Cohere’s work with individual companies is already “demonstrating a major avenue of growth” for the company, Gomez wrote. It has teamed up with Texas-based software giant Oracle to build custom AI models and power dozens of features across Oracle’s apps, Gomez wrote. Cohere’s partnership with Japanese information technology company Fujitsu has produced a multilingual model Takane, which specializes in Japanese.

The company plans to announce further partnerships in the coming weeks and says it will offer companies “AI capabilities that a generic consumer chatbot never could,” Gomez wrote.

Cohere is eyeing companies that will help them create models beyond English, co-founder Nick Frosst said in an interview with The Logic in November. Most LLMs have been focused on English so far, he added. 

“We’re very interested in making models good at all languages,” Frosst said. “The partnership model that we had with other companies has worked really well. We’d be interested in supporting other companies to do the same.”