Building intelligent enterprises: why developers must bridge code and business reality

In recent years, software engineering within large enterprises has undergone a significant transformation. Traditionally, business teams would outline requirements, and centralized IT units would build the software. However, the emergence of generative AI (Gen AI), low-code tools, and agentic workflows is reshaping this model. Subish Ram, Digital Products and Services leader at EY Global Delivery Services, highlighted this fundamental structural reset during a talk at DevSparks Hyderabad.

Over the past decade, EY has automated various processes across its global organization, supporting over 400,000 employees across its member firms. Ram emphasized that engineering teams can no longer remain isolated in technical silos. "The standalone IT team that solely focuses on software development is becoming less prominent," he observed.

"Instead, there is a growing shift toward placing engineering teams within the business side of the house. We require individuals who understand the business, can communicate in the language of business, and can integrate technology effectively." The shift from centralized IT to business prototyping began nearly a decade ago, with robotic process automation (RPA) addressing repetitive back-office tasks.

As RPA evolved, it progressed into intelligent workflows and now advanced to full-fledged agentic AI systems capable of orchestrating end-to-end business functions, such as finance, procurement, and human resources. With tools like Replit, Cursor, and enterprise-approved copilots becoming readily accessible, non-technical teams are increasingly taking the lead in development.

Tax consultants, HR leaders, and finance analysts are now building initial versions of internal tools on their own. Ram noted, "The initial stage of development is now taking place within the business environment. The technology team is then asked to consider scaling the solution, deploying it, and conducting security testing." This transition changes the developer's primary responsibility.

In the near future, business units will be responsible for basic prototyping, testing, and AI-assisted bug resolution. On the other hand, the engineering community will handle system architecture, integration, governance, and enterprise-grade scalability. Regarding the enterprise ROI and cost equation, Ram provided a realistic perspective on AI adoption challenges.

Unlike traditional RPA, where replacing manual hours directly led to cost savings, AI implementations involve significant recurring costs related to token consumption, infrastructure, and maintenance. "Calculating clear return on investment (ROI) remains a significant hurdle right now," Ram explained. "While automating a complete reporting process using AI may offer benefits, the continuous costs, such as hundreds of thousands of dollars annually for token usage, often result in no immediate ROI.

We have observed business cases where token costs have even surpassed human labor costs." Ram stressed that end customers ultimately prioritize productivity, efficiency, and the bottom line, rather than the underlying technology used—be it Python, traditional dashboards, or autonomous agents. To create sustainable value, developers must focus on developing reusable accelerators, enhancing token economics, and reducing AI consumption costs.

In the AI-native world, domain context has emerged as the new technical moat. To remain indispensable, engineers do not need to become domain experts, but they must understand the domain's mechanics. "You don't need to become a chartered accountant to build a finance solution, but you must understand the language used by accountants and the business outcome they seek," Ram concluded.

As AI tools simplify code generation, developers' key competitive advantage will lie in their ability to understand the business problems they are solving.