Leadership Spotlight
Humans Will Keep Defining the Problems That Matter with Nitin Narkhede
AI can accelerate the solution, but humans still have to define the problem. As machines take on what is already known, our greatest value will come from deciding what should be solved next.
A Conversation with Nitin Narkhede, Chief Technology Officer
Artificial intelligence is often discussed as a race for better models, agentic automation, and greater computing power. For Nitin, that framing overlooks the more difficult challenge: translating the knowledge of an enterprise into technology that can operate reliably in the real world.
In this conversation, he explains why AI must go beyond efficiency, how domain knowledge becomes part of an AI system, and why collaboration between engineers and subject-matter experts will become more important as development accelerates.
You have experienced several major technology shifts during your career. What separates a technology that creates genuine change from one that simply generates attention?
Two things have to come together: the potential of the technology and its adoption.
A technology can be extraordinarily capable, but if people and enterprises do not adopt it, it will not transform anything. We have seen promising technologies like mixed reality and blockchain that never fully crossed into the enterprise and technologies like metaverse that did not translate into people’s everyday lives.
With AI, the potential is clearly there. But having lived through earlier technology shifts, I have learned that the technologies that endure are the ones where practical adoption follows the initial hype. They fit into the reality of how people work, and then gradually change that reality.
Many companies say they have domain-led AI. What does that mean in practical, technical terms?
We often talk about domain knowledge as though it were a single body of information. In reality, it has at least four components.
- The first is declarative knowledge: the facts, concepts, terminology, and principles we learn through formal education.
- The second is procedural knowledge: the processes, methodologies, and skills we acquire through work.
- The third is conditional knowledge: knowing which rules apply, how context changes a decision, and when an exception needs to be made.
- The fourth is experience-based knowledge: the pattern recognition, intuition, and mental models that experienced professionals develop over time.
When we build an AI solution, these different forms of knowledge must be translated differently. Declarative knowledge can become taxonomies, ontologies, and knowledge graphs. Procedural knowledge can become workflows and AI agents. Conditional knowledge can be represented through business rules, policies, algorithms, and guardrails.
Experience-based knowledge is harder to encode. Much of it remains with the human expert. That is why our approach is not to separate AI from people. We build agents that work with experts so that they become significantly more productive.
How does that approach translate into a distinctive capability for Evalueserve?
Evalueserve brings together four essential layers: data, domain knowledge, engineering, and AI. The strength is not in treating them as independent capabilities but bringing them together around the client’s problem.
We are also thinking carefully about how intelligence can be reused across different domains. Consider counterparty risk, credit memos, and supplier risk. These are use cases from three very different domains, but each may need to understand a company’s financial performance, regulatory or legal exposure, sanctions status, creditworthiness, and early-warning signals. Instead of rebuilding this intelligence every time, we can create reusable components in a shared foundation layer. The foundation contains capabilities that apply across many problems. Above that, we add increasingly specialized knowledge, rules, and expertise for the particular use case.
That architecture gives us reuse, but it also makes the solution more robust. At the most specialized levels, our subject-matter experts bring the judgment required to interpret the situation appropriately.
Organizations are under pressure to demonstrate that they are “using AI.” What are they most likely to overlook?
Business and IT teams can become too focused on the visible component, such as incorporating an LLM into a solution, rather than asking whether it is genuinely the best technology for the problem.
We have come across instances where an existing software solution performs a task better, but the customer still asks for an LLM because leadership wants to see evidence that the organization is adopting AI. The motivation is understandable. But adoption without the right foundation can create the appearance of progress without delivering the expected outcome.
Data is part of that foundation, but enterprise data is much more than a collection of database records. Every organization has its own customers, offerings, historical work, case studies, processes, terminology, and internal language.
AI may know the wider domain, but the specifics of an enterprise live in its data, processes, and people. An effective system has to understand that context and operate within the organization’s rules.
If AI can generate software, will organizations eventually require fewer engineers?
I believe we will build more software as AI makes it faster and cheaper to develop it. As a result, we will need more engineers, although what they do will change significantly.
Instead of spending all their time at a keyboard writing code, an engineer may direct a group of AI agents. Those agents will take the engineer’s ideas and guidance and turn them into software. The engineer will define intent, establish boundaries, evaluate the output, and design strategies for testing what AI has created.
We saw something similar with cloud computing. Before the cloud, services and systems-integration companies performed massive amounts of infrastructure-management work. As cloud providers made infrastructure easier to manage, that category of work declined. But those companies did not shrink. New demand emerged around implementing software-as-a-service products, integrating systems, cloud migrations, and helping organizations use the new cloud infrastructure.
For Nitin Narkhede, the next stage of enterprise AI will not be defined by who deploys the greatest number of AI applications or agents. It will be defined by who can translate organizational knowledge into dependable systems, reuse intelligence across business processes, and combine automation with expert judgment.
That perspective reflects Evalueserve’s larger capability: uniting data, domain expertise, engineering, and AI to build solutions that understand not only an industry, but the distinctive language, processes, constraints, and opportunities of each enterprise.
Latest Posts
Stop Calling Them “Seniors”: The Food Opportunity Hiding in Plain Sight
Healthy aging needs are showing up across the wider food market, not only among older shoppers. This article covers four value pools: preventive nutrition, functional formats, convenience and premium trust. It also lists the evidence food companies need before they invest.
A Practical Guide to Building Agents That Survive Beyond the Demo
Building an agent is easy. Keeping it working is the hard part. Find out why reliability, well-chosen context, earned autonomy and trust are what make agents succeed beyond the demo.