Decoding the agentic future: An opportunity for Indian enterprises

Alexy Thomas, Partner, Technology, EY India

Most big Indian enterprises are well into their AI journeys. Many used ML for some functions before Generative AI and Large Language Models (LLMs) burst into the scene three years ago. However, most enterprises now have GenAI pilot projects and are also experimenting with GenAI chatbots. In large organisations, co-pilots are now an intrinsic part of most popular enterprise software.

The real AI revolution, as far as enterprises are concerned, though may take place next year – when the AI agents that all companies from Microsoft to Salesforce and Open AI are working on, become available for clients. In fact, while the long-term Holy Grail of all the big AI companies remains Artificial General Intelligence (AGI), the short-term goal is to take the lead in the race for an agentic future.

But how are AI agents different from the current generation of co-pilots and why are they expected to make so much difference? While both co-pilots and agents extend the LLMs functionalities, the primary difference is at the level of human interaction required. Co-pilots are powerful LLM tools – but they work alongside the user and need human interaction. They essentially speed up users’ work multi-fold, allowing them to utilise the full power of the software and reduce the number of steps that the pre-co-pilot software required.

On the other hand, agents can be autonomous and work like a team mate to a large extent – they can be trained to complete work based on the company’s procedures, goals and context without necessarily needing constant human oversight or supervision. The co-pilot, while powerful and adding to the enterprise software’s functionality, needs direction and unleashes its full power only when the user knows how to utilise it properly.

While many AI companies plan to release AI agents, they are not ready to function straight out of the box. The agents need to be configured and trained. By leveraging properly trained, sector-specific enterprise data, domain-specific agents can be developed by integrating agentic features into AI models tailored for areas like ESG, finance, and sales, ensuring enhanced performance.

This is where Indian companies need to understand the importance of data preparedness for an agentic future. Most enterprises – in India and globally – are at a lower stage of data maturity than is required for taking advantage of AI agents. While all AI relies on data and data management in Indian enterprises has evolved over the years, moving from data warehouses to data fabrics and lakes in the cloud, and companies have learnt to harness both structured and unstructured data to a large extent, this is not enough. What is needed for a true agentic future is what can be termed ‘Data 4.0’, which means, the enterprise data is trusted, catalogued, governed using AI for data to automate governance and has a real time data architecture.

For Data 4.0, organisations need to align both their structured and unstructured data with their core AI objectives. For example, a bank or a non-banking financial company requires three types of data to assess a loan application: Structured data (financial history, tax returns, credit scores, cash flow statements, gross debt, net debt, etc.); unstructured data (sales data trends, changes in external market conditions, economic data trends); and data on currently relevant lending regulations.

A co-pilot would help a human lender to analyse the structured and unstructured data and help in decision making. But an AI agent can be trained to decide on the loan application by analysing the dynamic data and taking decisions based on pre-set rules that take into account regulations as well as risk and prudential norms.

In a different setting – a manufacturing unit, for example – the agent can be trained for a very different task. It can be taught to decide when to slow down or even stop a production line, based on a constant stream of real-time data on quality of output, inventory levels, supply side bottlenecks, etc.

Data 3.0 also uses “Big Data” but is fundamentally focused on foundational tasks such as adding metadata and tags, adding fields, and preparing data for ML algorithms. Data 4.0 is a data-first approach. It is a cloud-native, metadata-driven paradigm that treats data as a central, strategic asset and designs the AI strategy around it.

An agentic architecture in the Data 4.0 framework would include a knowledge layer to capture organisational knowledge using knowledge graphs and vector database embeddings. This layer enhances agents’ capabilities by enabling them to understand processes, handle exceptions with human-in-the-loop support, and learn to act independently in future scenarios based on human input.

The agentic future holds a lot of promise, provided enterprises can take advantage of it. Indian organisations need to shift gears from a Data 3.0 mindset to a Data 4.0 framework to prepare for the future.