The Great Refactoring & How Cohere’s CEO is Rethinking Enterprise AI

I I have been following Cohere for over a year. Through reading interviews, participating in their Discord community, and working with their products, what strikes me most about Aidan Gomez isn’t just his impressive background—though co-authoring “Attention is All You Need” at Google Brain as an undergraduate certainly qualifies. It’s his refreshing clarity about artificial intelligence. In an era where AI discourse often swings between utopian hype and apocalyptic doom, Gomez stands out for his measured, practical perspective.

Having worked alongside AI pioneers like Jeff Hinton (now a Nobel laureate) and collaborated with leading researchers at Google Brain, Gomez could easily lean into the more sensational aspects of AI development. Instead, he focuses on something more fundamental: how to make this technology truly useful for businesses today.

This grounded approach makes Gomez particularly believable when he talks about enterprise AI. He doesn’t deal in hypotheticals or far-future scenarios. Instead, he discusses concrete problems and practical solutions, informed by actual customer experiences and market realities. This clear-eyed perspective, combined with Cohere’s focus on enterprise applications, positions them uniquely in the AI landscape.

When Aiden Gomez describes artificial intelligence, he doesn’t talk about sentient robots or existential risks. Instead, he speaks about concrete and supercomputers – quite literally. “It’s kind of like we have to repave every road on the planet,” he states, “and there’s like four or five companies that know how to make concrete.”

As the co-founder and CEO of Cohere, valued at over $5 billion, Gomez has an unusually grounded perspective on AI’s future. This shouldn’t be surprising – before founding Cohere in 2019, he was a co-author on “Attention is All You Need,” the landmark 2017 paper that introduced the Transformer architecture powering today’s AI revolution.

The Path to Enterprise AI

I’ve learned that Cohere’s mission isn’t to build the next ChatGPT. Instead, they focus on helping enterprises adopt and integrate AI technology. This means creating practical tools like research assistants for insurance actuaries responding to RFPs, or systems that help doctors quickly analyze decades of patient medical records before appointments.

“What we want to build is a platform and series of products to enable enterprises to adopt this technology and make it valuable,” Gomez explains. This focus on practical application over theoretical possibilities shapes everything about how Cohere approaches the market.

Enterprise Adoption: Learning from Early Mistakes

What’s particularly interesting about Cohere’s approach is their deep understanding of where enterprises typically stumble. Gomez notes that 2023 saw numerous proof-of-concept projects fail, often due to fundamental misunderstandings about how to implement AI effectively.

“All language models are quite sensitive to prompts, to the way that you present data,” Gomez explains. He points out that enterprises often overestimate the models’ capabilities, treating them like humans rather than understanding their idiosyncrasies. This is particularly evident in RAG (Retrieval-Augmented Generation) systems, where how you present retrieved results to the model can make or break the system’s effectiveness.

Based on my research, Cohere has developed a two-pronged approach to these challenges. First, they’re making their models more robust to handle diverse data presentation methods. Second, they’re developing structured products with clear APIs that guide users toward successful implementations, moving beyond simple raw model access.

Real-World Applications

The range of enterprise applications Gomez describes is remarkably broad. He highlights several key use cases:

1. Manufacturing Q&A: Allowing assembly line workers to chat with a system that understands all their tool manuals and diagnostic guides.
2. Enterprise Knowledge Access: Creating centralized chat interfaces for employees to access IT FAQs and HR documentation.
3. Healthcare Records Analysis: Processing decades of patient history to provide doctors with contextual briefings before appointments.
4. Insurance Underwriting: Building specialized research assistants that help actuaries rapidly analyze RFPs and respond more quickly.

The Build vs. Buy Decision

Gomez offers an interesting framework for how enterprises should think about AI implementation. He describes it as a pyramid: at the bottom are general needs that every organization shares (like employee chatbots), while the top represents highly specific applications unique to each business.

“As you push up that pyramid, it’s much less likely you’re going to find an off-the-shelf solution,” he explains. His advice? Don’t build what you can buy, and focus your development efforts on the unique applications that give you competitive advantage.

The Hidden Economics of AI

One of Gomez’s most interesting insights concerns what he sees as a fundamental misunderstanding about the AI market. While many observers talk about the “commoditization” of AI models, pointing to falling prices and open-source alternatives, Gomez sees something different: price dumping.

“You see people giving it out for free, giving it out at a loss, giving out at zero margin,” he explains. “They see the prices coming down and they assume prices coming down means commoditization.” But this state of affairs, he argues, is inherently unstable. The technology remains expensive to develop and deploy, and market pressures will eventually force a correction.

The Next Two Years: A Vision for AI’s Evolution

Looking ahead, Gomez sees several key developments on the horizon. He believes we’re entering a period where the core technology’s capabilities will continue to improve, but at a more measured pace. The focus will shift to integration and implementation.

“Even if we didn’t train a single new language model… there’s a half decade of work to go integrate this into the economy,” he explains. He predicts it will take another two to three years before developers become truly familiar with building these systems effectively.

The emergence of reasoning models represents another significant shift. These models will be able to work through problems step by step, catch their own mistakes, and iterate on solutions – capabilities that were previously limited. This development changes the economics of AI deployment: instead of waiting months for a new, more capable model, enterprises can allocate more compute time for better results.

The AGI Question

When I ask about artificial general intelligence (AGI), Gomez’s response is characteristically pragmatic. “AGI means a lot of things to a lot of different people,” he says. While he believes in building generally intelligent machines, he sees it as a continuous progression rather than a sudden breakthrough.

More importantly, he doesn’t believe we need to wait for some hypothetical superintelligent AI to create value. “We’ll be the ones to create abundance,” he says. “We don’t need to wait for this God to emerge and do it for us.”

The Great Refactoring

When Gomez talks about AI’s impact on business, he uses an analogy that cuts through the typical hype: “There’s a total technological refactor that’s going on right now and will last the next 10 to 15 years,” he explains. “It’s kind of like we have to repave every road on the planet and there’s like four or five companies that know how to make concrete.”

This metaphor is particularly apt. Just as roads form the fundamental infrastructure of physical commerce, language models and AI systems are becoming the essential infrastructure of modern business. The “repaving” isn’t just about replacing old technology with new – it’s about fundamentally reimagining how business processes and systems work.

The timeline Gomez suggests – 10 to 15 years – speaks to both the magnitude of the transformation and its inevitability. This isn’t a sudden disruption but a methodical rebuilding of business infrastructure. And just as you can’t repave all roads simultaneously, organizations will need to strategically plan their AI transformation.

His point about the limited number of companies that “know how to make concrete” is equally telling. While there’s much discussion about the democratization of AI, the reality is that building reliable, scalable, enterprise-grade AI systems requires deep expertise and significant resources. This reality shapes how organizations should think about their AI strategy – knowing when to partner with experts like Cohere versus trying to build everything in-house.

A Clear Path Forward

For business leaders navigating the complex landscape of enterprise AI, Gomez and Cohere offer something invaluable: clarity. While competitors chase headlines with consumer chatbots or make grand promises about artificial general intelligence, Cohere is focused on solving real business problems today.

Their approach – emphasizing security, reliability, and practical implementation – suggests why major enterprises are increasingly turning to Cohere for their AI needs. The company’s ability to deploy models in whatever environment customers require (on-premise, in VPC, or outside VPC) demonstrates their deep understanding of enterprise requirements.

For those looking to implement AI solutions in their organizations, Gomez’s insights align with what I’ve been telling customers—they offer a valuable roadmap. Start with the basics, focus on specific business value, and build from there. Don’t get distracted by the hype or discouraged by early setbacks. Instead, follow Cohere’s lead: focus on practical applications, understand the technology’s real capabilities and limitations, and build for long-term success.

As we enter what promises to be a transformative period in enterprise computing, leaders would do well to pay attention to Gomez and Cohere. Their grounded approach to AI implementation, combined with their deep technical expertise and focus on enterprise needs, positions them as a potentially crucial partner for organizations looking to successfully navigate the AI revolution.

The question isn’t whether AI will transform enterprise computing – it’s how quickly and effectively organizations can adopt and integrate these new capabilities. In Cohere, we see a company that isn’t just talking about this transformation, but actively making it happen, one enterprise at a time. For business leaders serious about AI implementation, that makes them worth watching very closely indeed.