Adobe, a global leader in creativity and digital experience solutions, continues to push the boundaries of how data, analytics, and AI drive business transformation. With products like Photoshop, Acrobat, and Adobe Experience Cloud, the company serves millions of users worldwide and generated $21.51 billion in revenue in 2024. At the core of this innovation is Adobe’s enterprise-wide commitment to building a scalable, AI-first data infrastructure that powers everything from customer insights to personalized experiences.

This article marks the final part of a three-part series featuring Bin Mu, Chief Data Officer and VP of Enterprise Data & Analytics at Adobe, in conversation with Rohit Choudhary, CEO of Acceldata. The first installment explored the evolution of Adobe’s data and AI strategies. The second part examined how Adobe is redefining the semantic layer, rethinking dashboards, and delivering real-time, AI-driven intelligence directly to users.

In this concluding discussion, Mu shares how Adobe’s technology stack is evolving, the company’s approach to AI models, and the ways AI is reshaping organizational structure and leadership.

Edited Excerpts

Q: Before we dive into organizational structure, could you walk us through how Adobe’s technology stack is evolving, especially with your AI-first approach?

First, everything has to be in the cloud. For any company, if the data or analytics platform is still on-prem, there will be challenges. Second, you need a strong foundation. For us, that starts with a core data intelligence platform.

To centralize data access, you also need a data integration platform that can pull together different source and transactional systems, so data can flow seamlessly through the intelligence layer for AI and analytics. We’ve also built a reference data platform for marketers and external users, enabling data enrichment, enhancement, and validation through third-party providers.

Then there’s the delivery layer — whether through a fabric like Microsoft Fabric or a native delivery mechanism. At every stage, governance and quality controls are critical. Data lifecycle and management considerations have to be built in from the start.

On top of that, metadata management is a core function, with cataloging at the center. Many tools can support this, but the process itself must be treated as foundational.

Finally, with Agent AI and AI orchestration, the possibilities are limitless. You can have hundreds of agents handling various tasks, but what really matters is how you structure the network and orchestrate their interaction. That includes managing agent registration, so what we build doesn’t get lost in the process. These pillars form the backbone of how we’re thinking about our stack.

Q: You mentioned infrastructure and the data supply chain. What about models? Are you leaning toward open-source, commercial offerings, or building your own?

There isn’t a “right” or “wrong” approach — it’s about what fits the need. We love open source because of the community contributions, but we always pair that with enterprise support. We also build our own AI models. Adobe has its Firefly AI model, for example. We’ve built other models too, but decommissioned a couple because the investment didn’t make sense compared to using models from providers like OpenAI or Gemini.

Our agent network is model-agnostic. You can swap in or out any model. We treat AI models as commodities, which gives us flexibility and speed.

Q: With this transformation, people often worry about job displacement. How do you see AI reshaping organizational roles and leadership at Adobe?

It’s a transformation, not just a shift. Roles will evolve. Some positions will disappear, but at the same time, new opportunities will emerge. Every technological revolution has done this — it changes professions and, ultimately, drives progress.

Take the 1970s and early ’80s. Telephone switchboards employed hundreds of thousands of people, plugging in calls manually. Today, that’s all automated, and we’ve created entirely new industries since then. AI is no different — it will reshape roles and open up new possibilities.

As leaders, we have to be fluent in technology. I tell my own leaders: “Be hands-on with AI.” Use personal AI to write code, build a conversion tool, and explore what it can do. That hands-on experience helps you think clearly about implications for the organization and the opportunities to work differently.

One example is requirement gathering. Today, translating business requirements into technical specifications is often a painful process. With AI, we’re experimenting with using copilots to record conversations, then having agents analyze past Jira registries, ask clarifying questions, and even suggest requirements. We’re not fully there yet, but the potential is real.

On a personal level, I use AI every night to summarize my day, organize my tasks for tomorrow, and track progress over time. That daily interaction gives me insights that help me lead better.

Looking forward, we’re going to center our organizational structure around AI. The goal is to harness AI’s full power across analytics and operations — not as a bolt-on, but as a core capability shaping how we work and deliver value.

CDO Magazine appreciates Bin Mu for sharing his insights with our global community.