AI agents are moving beyond basic automation and reactive interactions to support more complex enterprise workflows. Advances in foundational models and computing power are enabling AI systems to process large volumes of information, interact with multiple data sources, and execute tasks within defined boundaries.
For enterprises, this shift is particularly relevant because employees often spend significant time collecting information from different applications, processing data, and following established workflows. Agentic AI can connect these activities and support faster execution while keeping human review within the process where required.
In this episode of the Analytics Insight Podcast, Priya Dialani speaks with Vikram Jeet Singh, Partner at BTG Advaya, to examine how agentic AI is changing enterprise workflows, where organizations can apply these systems, and what businesses need to prepare before deploying them at scale. Here are the excerpts of the interview:
I see generative AI as largely reactive. We ask it a question or give it a prompt, and it provides an answer. It is extremely useful for creating content, summarizing documents, answering questions, and supporting productivity.
Agentic AI adds another dimension. I see it as an extension of generative AI where the system understands an objective, plans the required actions, interacts with different information sources, and completes multiple tasks within defined boundaries.
The system can work with databases and enterprise applications, make decisions based on established rules, and execute one step after another. It can also be designed to return to a human or analyst for confirmation when necessary. That is what makes agentic AI more applicable to complex enterprise workflows.
I believe the greatest impact will come in areas that require large amounts of information to be processed and analyzed. In many organizations, employees have to collect information from 15 or 20 different applications before they can complete a task.
Research and audit are good examples. At Decimal Point Analytics, we have built systems that source information from company filings and calendars, identify quarterly results, download transcripts, and summarize the information. An analyst can then review the output before it is released to customers.
I also see strong applications in risk and compliance. Agents can monitor regulations, compare them with internal policies, identify gaps, and alert compliance teams. Customer service, sales, business development, accounting, and reconciliation can also benefit from these workflows.
I would start with enterprise knowledge. An agent needs context to understand how the organization operates. That knowledge may exist in emails, standard operating procedures, internal documents, or even in the experience of managers and employees.
We need to bring that knowledge into structured repositories so the system can use it. I consider this knowledge base an important part of the organization's intellectual property.
The second requirement is data. We need efficient pipelines that allow agents to access information from ERP, CRM, accounting, HR, and other enterprise systems. We also need clearly defined workflows and standard operating procedures. These foundations are particularly important when agentic AI is introduced into time-sensitive or mission-critical processes.
I think the challenges are closely connected to the foundations required for deployment. Organizations first need to ask whether their data is properly structured and whether their workflows and standard operating procedures are clearly defined.
A lot of important knowledge may still exist outside formal databases. It may need to be organized and converted into a form that an AI system can use effectively.
There are also concerns around confidentiality. Organizations need to understand how their data will be handled and whether sensitive information could be exposed beyond the intended environment. Cost is another consideration because enterprises need to evaluate the economics of operating these systems alongside their technical and data requirements.
I believe organizations should start considering agentic AI as early as possible because the technology is moving quickly and adoption is already taking place across different organizations.
At the same time, I would not treat deployment as simply adding another software system. The enterprise needs the right knowledge base, data pipelines, workflows, and operating structure first.
Once those pillars are established, agentic systems can be introduced into business processes where they can handle large amounts of information and time-sensitive work. I see the opportunity as moving beyond task automation toward systems that can coordinate multiple steps while still allowing people to review and guide important decisions.
Listen to the full discussion on the Analytics Insight Podcast.