The Shift to AI-Native Is About Reinventing the Enterprise, Not Just Adding Tech: CEO of EPAM

The Shift to AI-Native Is About Reinventing the Enterprise, Not Just Adding Tech: CEO of EPAM

To truly realize value from AI, organizations must rethink processes across every function: CEO of EPAM

As enterprises move beyond AI pilots, the challenge is no longer deploying AI tools but redesigning processes, operating models, talent, and economics around them. In an interview with ET CIO, Balazs Fejes, President and CEO of EPAM, discusses what it really takes to become AI-native, why AI literacy should precede transformation, how token economics is reshaping discussions, and why governance must evolve alongside innovation.

Q. Enterprises are increasingly talking about becoming AI-native rather than simply adopting AI. What does that transition really mean?

FB: Many organizations initially believed AI adoption meant purchasing AI licenses, rolling out copilots and training employees. But becoming AI-native is fundamentally different. It is about changing how the enterprise operates, not just introducing new technology. 

To truly realize value from AI, organizations must rethink processes across every function. This includes how they go to market, how sales and account teams engage customers, how HR and finance operate, and how delivery teams embed AI into day-to-day execution. 

Most organizations still carry technical debt and organizational complexity. Becoming AI-native requires systematically redesigning those processes across departments and functions. It's a multi-year transformation rather than a technology deployment.

Engagements themselves also have to mature. In the beginning, adoption was ad hoc: people installed the software and used it when they felt like it. What we are moving towards is what we internally call “Level Three:” AI embedded into the delivery system itself, in a genuinely disruptive way, rather than layered on top of the existing way of working.

Q. What makes this transformation difficult for enterprises?

FB: It’s inherently complex. Early in any transformation, enthusiasm comes naturally because early adopters embrace change. But sustaining that momentum is much harder.

Organizations need champions within every function who continuously drive adoption, customize AI for different business teams and keep employees engaged. AI cannot simply be deployed as another software tool; it has to become part of how work gets done.

Another challenge is organizational fatigue. Employees are already dealing with continuous change, and AI adds another major transformation. Companies, therefore, need structured adoption programs while simultaneously preparing fresh graduates entering the workforce. Delivery organizations tend to move first, because they sit closest to the client. But you very quickly find that delivering with AI forces you to change everything around it: how you interview, how you run HR processes or how you budget.

The greatest risk is delaying transformation. Eventually, the gap between an organization and the AI frontier becomes so wide that transformation becomes harder. Beyond a point, it becomes easier to start a new company than to transform the existing one, and the window in which you can realistically do this is limited.

Q. What should an organization prioritize to become an AI-native company?

FB: The first priority should be AI literacy.

Before discussing advanced concepts such as token economics or AI governance, employees need a foundational understanding of AI. Organizations should invest in educating people across functions so they can meaningfully participate in AI-driven decision-making.

The board also needs to become AI-literate. Without board-level understanding, discussions around budgets, investments and business outcomes become difficult. On the work itself, I would start where cost pressure is highest and the processes are most repeatable — call centers, business process outsourcing and IT operations.

After building literacy, organizations should experiment internally. At EPAM, we first applied AI to our own IT and product development teams before extending those learnings to client engagements. We call this a "client zero" mindset — using your own organization as the first proving ground before scaling externally.

Only after building internal confidence should enterprises expand AI adoption to customer-facing operations.

Q. AI investments often face scrutiny around returns. Has demonstrating ROI become easier?

FB: There was a period when demonstrating ROI was relatively straightforward because AI models had matured while pricing remained subscription-based. Organizations could automate customer service, marketing and other functions with clear financial benefits. We had one client where we fully automated call center operations handling hundreds of thousands of calls, and another where we automated marketing. For six or seven months, ROI was straightforward if you were good at engineering. That has changed.

Today, achieving ROI depends on how intelligently organizations manage token usage, model selection and inference costs. AI economics has become significantly more sophisticated.

Enterprises now need to decide which workloads justify premium frontier models, which can run on open-source alternatives and where automation genuinely creates value. These decisions directly influence financial outcomes.

ROI still exists, but accessing it requires engineering discipline, careful architectural choices and an understanding of token economics rather than simply deploying AI models.

Q. You mentioned token economics. Do you expect it to become a mainstream enterprise concern?

FB: Absolutely.

Many organizations are only beginning to understand how token consumption affects AI costs. By the 2027 budget cycle, tokens will be a significant line item. But you need a finance function that can actually hold that conversation. If it cannot, the adoption will not happen either. That means finance leaders, technology leaders and business teams must develop a shared understanding of AI economics. Organizations that cannot have informed discussions around token usage, model costs and optimization will struggle to scale AI sustainably.

Q. How is AI changing workforce development and talent strategy?

FB: AI-native talent is arriving much faster than many organizations realize.

Graduates entering the workforce today have grown up using AI tools throughout their education. They approach problem-solving differently, experiment more freely and are comfortable working with AI models, coding assistants and autonomous agents.

We've seen new graduates build AI-native solutions using swarms of agents and modern coding models, applied both to building solutions and to operating them. That level of fluency would not have come out of a graduating class a couple of years ago. We are also running certification programs with Anthropic and Google, and a striking number of our juniors are stepping up to them.

This changes how organizations should think about junior talent. Rather than viewing them simply as inexperienced employees, companies should see them as catalysts for AI-native ways of working.

Equally important is helping existing employees adapt. Mid-level managers often represent the most resistant layer during transformation because they must change long-established work habits. Organizations need to actively support this group through education and continuous learning.

Q. What role should governance play as organizations become AI-native?

FB: Governance must evolve alongside AI adoption rather than follow it. The lesson from data governance is that it only works when you move it out of committees and into the foundations — into data contracts and data products, as a programmatic solution rather than a review board. AI governance has to go the same way.

Governance is no longer just about compliance. It now runs through an AI orchestration layer: where inference can and cannot be executed, how tokenomics and the financial operations around it is managed, how you contain the blast radius of agents, address cyber risk and monitor model drift and output quality. It is reminiscent of the early days of outsourcing, when every conversation was about whether a workload could go offshore. The question now is which models you are permitted to run, where and under what restrictions. The challenge is that AI innovation is moving faster than governance frameworks. Technology providers are building orchestration platforms, but enterprises must simultaneously establish their own policies around model usage, data access and AI deployment.

Q. Many organizations believe AI will reduce the need for human effort. Do you agree?

FB: I see the opposite happening.

AI reduces repetitive work, but it increases the need for engineering, experimentation and thoughtful decision-making. With AI, organizations have greater freedom to innovate, but they also need stronger governance. Once you take humans out of the loop, you need to document more, not less. The tacit knowledge that sits in people's heads, or in the notes layered on top of processes that have drifted over years, now has to be recaptured and written down, because agents cannot infer it.

In many ways, AI is exposing weaknesses that already existed within enterprises, particularly around data quality, documentation and process maturity. Becoming AI-native, therefore, requires strengthening those foundations rather than simply deploying more AI models. We are now exposing data to AI through mechanisms such as MCP servers — data that was never structured or intended for machine consumption, and often with no semantic layer above it. Organizations that have hoarded data for decades are handing it to AI and discovering it is not usable in that state. This is not a problem only for GCCs or service providers to solve; our clients have to fix these foundations, too.

View the original interview here.

Related reading: Why Engineering Talent Will Determine the Next Generation of AI Leaders

Discover how EPAM can help redefine your enterprise and outpace your competitors in the era of AI.