Chloe Maraina is a visionary leader in the realm of business intelligence, known for her ability to transform complex data architectures into compelling narratives that drive corporate strategy. With a deep background in data science and a specialized focus on the integration of emerging technologies, she has spent her career navigating the intersection of human expertise and machine intelligence. As enterprises move beyond the experimental phase of artificial intelligence, Chloe’s insights provide a roadmap for organizations looking to scale autonomous systems across global operations. In this conversation, we explore the strategic shifts occurring within the world’s largest corporations—from billion-dollar pharmaceutical partnerships to the psychological nuances of AI-driven customer service—and examine what it truly takes to move from a pilot program to a full-scale AI transformation.
The discussion highlights the evolving landscape of enterprise AI, focusing on the transition from surface-level experimentation to deep, strategic integration within massive organizations. Key themes include the critical role of human capital and embedded engineering talent in specialized fields like drug discovery, the importance of technical portability across the cloud stack to eliminate data silos, and the sophisticated engineering required to build trust through high-fidelity AI avatars. We also delve into the competitive dynamics between major cloud providers and the specific milestones necessary to train a massive global workforce for an agent-centric future.
Large pharmaceutical companies are increasingly moving past experimental AI to formalize massive, long-term partnerships. How do you assess the value of embedding forward-deployed engineers directly into research teams, and what specific milestones should a company target when training 75,000 employees on new agent platforms?
The value of embedding forward-deployed engineers, or FDEs, cannot be overstated because it bridges the gap between raw computational power and the nuanced reality of biological research. When a company like Merck signs a $1 billion deal, they aren’t just buying software; they are securing the presence of PhDs from organizations like DeepMind who sit across the table from their own scientists to solve the riddle of human biology. This “human capital” investment is often the deciding factor because having two or three consultants “moving the needle” simply isn’t enough for a multinational giant; you need a significant presence to actually accelerate drug discovery. For a workforce of 75,000 employees, the training milestones must move beyond basic literacy toward functional fluency in agent-based workflows across research, manufacturing, and commercial departments. Success is measured by how quickly these teams can transition from asking “what is this tool?” to “how can this agent automate this specific part of my research pipeline?” which ultimately gets life-saving medicines to patients much faster.
Organizations often struggle with data silos when deploying AI voice agents and customer experience tools. What are the practical advantages of maintaining portability across different layers of the tech stack—from databases to front-end frameworks—and how does this connectivity impact the speed of model training and deployment?
Maintaining portability across the stack is the secret to avoiding the “pilot purgatory” that many companies find themselves in. If you look at a major retailer like The Home Depot, their journey started over a decade ago by moving their website to the cloud, and now that foundation allows them to move seamlessly from BigQuery databases to Gemini Enterprise frameworks. This “connective tissue” allows data to flow without friction, meaning that an AI voice agent in a customer experience role has the same context as the backend inventory system. When the layers of the stack—from the database to the agent development kit—are integrated, the speed of deployment increases exponentially because you aren’t rebuilding the wheel for every new use case. This fluidity ensures that the AI isn’t just a bolt-on feature but a core part of the infrastructure that can scale as quickly as the business demands.
Many enterprise tech vendors now offer similar AI agent platforms featuring integrated knowledge graphs and governance security. In a market where core capabilities seem to overlap, what specific technical nuances or human-capital support models actually influence a major corporation’s decision to switch or consolidate providers?
We are currently seeing a surge in “minimally differentiated platforms” where the marketing language from one vendor sounds almost identical to the next. In this environment, the decision to consolidate often comes down to the depth of the partnership and the specialized expertise the vendor brings to the table. For instance, while many providers offer agent tools, Google’s ability to provide DeepMind expertise was a primary differentiator for Merck, providing a level of machine learning depth that others couldn’t match. Large enterprises are looking for partners who are willing to commit substantial human resources rather than just providing a license and a login. When the technical features like knowledge graphs and security protocols become table stakes, the “significant investment” in people and the willingness to co-innovate on complex problems like human biology are what truly move the needle for a C-suite executive.
Developing AI avatars requires high-quality video and voice output to establish trust with banking or retail clients. How does bypassing traditional voice-to-text translation improve the natural flow of these interactions, and what specific metrics determine if an AI agent has successfully built a “trusting relationship” with a user?
Bypassing the voice-to-text middleman is a massive technical leap because it removes the lag and the “robotic” cadence that usually breaks the illusion of a real conversation. For a project like Citi’s Sky AI avatar, the quality of the interaction is found in the smallest details, such as the natural inflection of a voice or the subtle movements during pauses between questions. Without this level of fidelity, the idea of a “trusting relationship” in a sensitive environment like wealth management would be a stretch for most customers. We measure success by monitoring user engagement and the emotional resonance of the interaction; if a client feels heard and understood without the friction of a delay, the agent has done its job. The ultimate metric is whether the user treats the avatar as a reliable partner rather than a frustrating barrier to a human representative.
Moving from small-scale pilot programs to a comprehensive AI strategy across manufacturing and corporate departments is a massive undertaking. Could you walk through the step-by-step process of updating business workflows to incorporate autonomous agents, and what common pitfalls typically derail these enterprise-wide rollouts?
The transition begins with a shift from “working around the edges” to defining a truly strategic, long-term partnership that touches every facet of the business, from R&D to commercial operations. The first step is embedding engineers directly into the business units to identify which workflows are ripe for automation, followed by the training of foundational models on the company’s specific data. A common pitfall is the failure to realize that AI agents are not just “chatbots” but sophisticated tools that require a complete rethink of how a department functions. Many rollouts fail because the organization doesn’t invest enough in the “human” side of the equation—failing to train the thousands of employees who will be interacting with these systems daily. When 75% of a cloud provider’s customers are already using AI, the companies that succeed are those that treat AI as a core business transformation rather than just another IT project.
What is your forecast for the evolution of enterprise AI agents?
I expect that we are moving toward a period of massive consolidation where the “billion-dollar deal” becomes the new standard for enterprise cloud partnerships. In 2025, we already saw more of these massive agreements than in the previous three years combined, signaling that the era of small-scale experimentation is officially over. We will see AI agents move from being passive assistants to active “co-workers” that handle complex, multi-step processes in real-time, such as managing a global supply chain or conducting early-stage drug synthesis. The focus will shift from the models themselves—since most providers now support multiple models—to the management side, specifically how these agents are governed and secured at a global scale. Ultimately, the winners will be the organizations that can successfully blend elite human expertise with these autonomous platforms to create a faster, more responsive version of their current selves.
