AI agents are moving beyond chatbots to power enterprise decisions

As organizations face more and more complex workflows, combined with increasingly high expectations for speed, accuracy and flexible adaptation, a new generation of automation is emerging — Agentic AI. “Agentic systems, unlike the standard chatbots that only respond to scripted user queries, can autonomously operate in context and purpose” states Mr. Jayachandran Ramachandran, Senior Vice President - AI Labs at C5i.

In this interview, he explains how AI agents are already solving a variety of real-world enterprise problems ranging from supply chain and HR, to molecule assessment in pharma, and details C5i's AgentHub, Xelerate.AI and AssetPro platforms that have the ability to deploy AI quickly, transparently and with trust.

Excerpts from an interview: 

Agentic AI is being called the next leap beyond chatbots. How do you define its enterprise relevance, and what real-world problems can it solve today that traditional automation couldn’t?

An AI agent is an autonomous software entity designed to perceive, reason, plan, and take actions to achieve specific goals. Unlike traditional chatbots, which are typically script-driven and react to predefined queries, agents possess true autonomy and goal-orientation. While a chatbot might answer simple queries, an agent will break down a complex problem, devise a plan, and execute actions across multiple systems to solve it. Agents can learn and adapt from experiences, remember context across interactions, and initiate actions.

This capability helps agents to automate enterprise workflows that are complex, involve interconnected systems, hold large volumes of structured and unstructured data, and numerous human touch points. Agents are solving many real-world problems such as in the marketing domain, agents that can continuously monitor market trends and competitor moves and proactively deliver strategic recommendations for new product features or pricing or promotions. In supply chain, agents can monitor inventory, identify supply disruptions and autonomously reroute shipments or readjust stock across warehouses or identify alternative suppliers.

Agents can collaborate with humans to bring in higher levels of automation for end-to-end workflows such as hire to retire, order to cash, procure to pay, idea to launch etc. Being goal oriented, agents can augment human to drive key organizational success metrics like revenue, efficiency, cost savings, enhanced customer satisfaction, and improved risk mitigation.

C5i has launched platforms like AgentHub and Discovery. What’s the biggest challenge in building GenAI products that actually work in complex enterprise environments?

The biggest challenge in building successful GenAI products for complex enterprise environments lies in bridging the gap between the current LLM capabilities/ its limitations and the stringent demands of real-world business operations.

While architecting GenAI products, it is important to consider the functional and non-functional requirements of the systems and identify the LLMs that can meet expectations around accuracy, reliability, cost, latency, security, privacy, explainability etc. For enterprise applications dealing with critical decisions such as finance, legal and healthcare, even a low error rate is unacceptable. For users to adopt GenAI tools, they must trust the output. 

A rigorous quality assurance process combined with strict adherence to guardrails is essential for establishing trust. Building transparency around how models arrive at decisions and the reasoning behind that enhances trust. A human-in-the-loop approach is vital as it helps to validate outputs, correct errors, and provide feedback for continuous improvement.

For successful adoption of GenAI products, it is important to train users, embed it in their daily workflows and enable habit-forming practices without creating any friction. It is equally important to have a clear path to ROI on GenAI initiatives by establishing baselines, measurement criteria, continuous monitoring to realize benefits.

With AssetPro, you’re applying AI to early-stage molecule assessment. How is applied AI reshaping the life sciences value chain—from discovery to market access?

In the pharmaceutical industry, the BD&L team (Business Development & Licensing) plays a critical role in shaping the company's strategy, future pipeline, and overall growth. Their primary objective is to identify, evaluate, negotiate, and execute strategic partnerships, alliances, acquisitions, and licensing agreements that align with the company's therapeutic areas of interest and long-term goals. The identification of a potentially successful molecule is key to any BD&L team. It involves spending weeks and months of effort sifting through enormous piles of public and proprietary data and cross examining them to arrive at top recommendations for the management. AssetPro completely automates this process through automated data harvesting, identifying relevant documents, assessing the progress of the assets in the clinical trials, examining the safety and efficacy parameters and their market potential. Based on preconfigured scoring criteria, AssetPro recommends the top assets which aligns with organizations strategic objectives thereby substantially improving BD&L team performance.

Enterprises are struggling to balance AI power with explainability. How does C5i ensure transparency, control, and ethical guardrails in your GenAI deployments?

C5i has a well-defined Responsible AI framework which focuses on ensuring AI models are reliable, safe and secure, privacy preserving, explainable and interpretable. The framework helps to understand the business use case in detail and assess the risks across dimensions such as data, AI model, usage scenario etc. Based on the risk profile, the AI system will be subject to different levels of checks as required. We have a well-established QA process to validate outputs from AI systems before they are deployed.

There are post implementation monitoring as well to ensure there is consistency in the outputs and there in no deviation from the acceptance criteria. While LLMs on cloud platforms provide some guardrails, these are further augmented with use case specific guardrails for domain adaptation. By leveraging the reasoning power of LLMs, we broadly understand the approach adopted for inference and use that indirectly to explain and interpret AI model outputs. This is an evolving space, and we are constantly on the lookout to explore and experiment to improve trust in AI outputs.

Speed to market is critical. How does Xelerate.AI help organizations reuse AI assets and rapidly scale innovation without reinventing the wheel each time?

C5i Xelerate is a platform of ready-to-use enterprise-grade AI models and reusable AI solutions. It helps to fast-track development of products and solutions with AI accelerators spanning generative AI, natural language processing, computer vision, machine learning, graph technology, reinforcement learning, data engineering and digital solutions with built-in Responsible AI and MLOps best practices. It’s the go to platform for C5i’s AI team to deliver solutions to our clients in the shortest time, with less effort and better quality. The platform is enriched on an ongoing basis with regular contributions from teams across the organisation. It helps to build and innovate at speed and scale thereby driving adoption and business impact.

What does the next evolution of AI look like in the enterprise? will we move toward fully autonomous decision systems, or is human-AI collaboration still key?

AI is driving enterprises to fundamentally reimagine business processes and associated roles that own them. The automation spectrum will vary from complete automation in low-risk/high-feasibility tasks to minimal or no automation in tasks that carry higher risks and low feasibility. The next evolution of AI in the enterprise is moving toward a multi-agent AI ecosystem where AI agents autonomously collaborate, optimize workflows, and make complex decisions with minimal human oversight.

With the arrival of new frameworks like MCP (Model Context Protocol), A2A (Agent to Agent), and ACP (Agent Communication Protocol), we are likely to witness the emergence of a larger, more interconnected ecosystem of AI agents that can seamlessly communicate, collaborate, and access diverse data sources and tools. 

However, human validation still remains crucial, especially given the limitations of large language models in accuracy, reliability, context understanding, and ethical judgment. The future will be hybrid project teams with a combination of agents and humans working together and supporting each other to deliver outcomes. This transformation requires new organizational structures, AI governance frameworks, and workforce reskilling to significantly enhance the likelihood of successful AI integration and value realization.

Looking across industries, which sectors are poised to benefit the most from agentic and applied AI in the next 2–3 years, and why?

The transformative potential of agentic and applied AI is sector agnostic and cuts across the enterprise functions. Every sector will see benefits, but the degree and nature of that benefit will vary based on current technology landscape, automation levels, data richness, complexity of their workflows, AI literacy and preparedness, culture and commitment of the organization.

For example, industry verticals such as Financial Services, IT and Managed Services, Retail and CPG, Technology and Telecom have currently higher levels of automation, and they are poised for deeper transformation. They will graduate from task level automation to full-fledged workflow automation and optimization. On the other hand, industry verticals like healthcare and life sciences, education, and manufacturing, which currently have relatively lower levels of automation, are well-positioned to leapfrog ahead and unlock significant benefits through accelerated adoption of Agentic AI.

They can embrace this innovative technology faster since they have less technical debt and are not constrained by the legacy rule-based automation systems.