Cohere Co-Founder Talks Proof-of-Concept Fatigue at Web Summit Vancouver
At Web Summit Vancouver, Ivan Zhang, co-founder of Cohere, addressed the growing proof-of-concept fatigue among enterprises eager to adopt AI technology but struggling to move beyond initial trials. During a panel discussion with Tailscale CEO Avery Pennarun, Zhang acknowledged the widespread frustration. “Everyone is tired of [proofs of concept],” he said. Many of Cohere’s clients, based in Toronto, have developed initial applications but have yet to deploy them in production. Challenges such as cost, governance, data security, and privacy are slowing progress, issues Cohere plans to tackle with its new workspace platform called North.
The Next Phase: Where’s the ROI?
In a follow-up interview, Zhang was clear about the current state of AI adoption: many companies haven't seen the return on investment (ROI) needed to justify their AI expenses. “Sometimes the systems they end up building, the cost of the model itself is more expensive than the humans that are actually running it,” he said. He also shared that attempts to boost workforce productivity with AI have sometimes fallen short, with humans doing less work but producing the same output. Zhang expects AI startups to focus on regaining trust from companies burned by past underperforming projects.
A National Bureau of Economic Research study surveying 7,000 workplaces found no significant impact of AI chatbots on earnings or hours worked. Similarly, a Boston Consulting Group report revealed that only 25% of executives surveyed have seen meaningful value from AI so far. The report suggests companies achieve better results by concentrating AI efforts rather than spreading them thin across multiple pilots.
Focus on Real Business Problems
Zhang advises companies to prioritize clear business problems before investing heavily in AI solutions. “Don’t get lost in building something and searching for a problem,” he said. He emphasized that AI and AI agents are just tools within a broader toolbox to solve business challenges and deliver customer value.
His remarks offer a grounded perspective amid ongoing industry hype about AI’s potential. NYU professor emeritus Gary Marcus, also speaking at the summit, criticized generative AI’s tendency to produce inaccurate or fabricated information, describing large language models (LLMs) as “auto-complete on steroids.” Marcus pointed out that increasing training data has not reduced hallucination rates, which remain stubbornly high even in the latest models from top companies.
Cohere’s Market Position
Cohere has made progress but still trails some competitors. According to the Hugging Face LLM hallucination leaderboard, Cohere’s newest Command A model reduces errors compared to previous versions but ranks 66th globally, behind several rivals. Zhang admitted that hallucination is an ongoing challenge. Cohere addresses this by increasing transparency, showing users “the raw thinking” of its models, including tools used and source citations.
After raising $500 million USD in Series D funding at a $5.5 billion valuation, bringing total funding to about $1 billion, Cohere faces tough competition from better-funded players. Zhang and others in the field argue bigger models aren’t always better. He said, “A model is only as good as the data and systems it can access.” Cohere’s products are designed to run fully within customers’ environments, pushing the company to develop more efficient models.
Zhang highlighted Cohere’s “intense growth” and noted the AI space remains young with significant room for expansion. He reiterated that AI tools serve to solve business problems and create customer value.
Revenue and Legal Challenges
Cohere recently hit $100 million USD in annualized revenue after doubling sales since early 2025. CEO Aidan Gomez told Bloomberg the company is “not far away” from profitability. However, reports indicate Cohere’s current revenue is still $350 million USD below the targets shared with investors in 2023.
Revenue and competition aren’t the only hurdles. Cohere faces a significant copyright-infringement lawsuit from major media companies like the Toronto Star, Condé Nast, and Vox. The plaintiffs allege Cohere scraped content without consent and used it to train AI models, generating infringing outputs. Cohere denies these claims, disputing the notion of copyright infringement and accusing the publishers of manufacturing a case. Zhang referred inquiries to a company blog post and expressed confidence in their position.
For IT professionals and developers interested in practical AI applications and challenges companies face, this discussion highlights the importance of focusing AI efforts on solving clear business problems and managing expectations around ROI and technical limitations.
Explore more about AI tools and training options at Complete AI Training.
Your membership also unlocks:
