Agentic AI: Autonomy, Problem-Solving, and the Future of AI Systems

In this insightful interview, Ramprakash Ramamoorthy, Director of AI Research at ManageEngine, speaks about the evolving landscape of artificial intelligence with a focus on Agentic AI. He defines how these autonomous systems differ from traditional AI, explores their unique ability to solve complex, multi-step problems in IT operations, and discusses the critical balance between AI autonomy and human oversight. 

How do you define Agentic AI, and how does it fundamentally differ from traditional chatbots or rule-based automation?
Agentic AI refers to AI systems capable of operating autonomously, making decisions, initiating actions, and adapting strategies in response to their environment. Unlike traditional AI, which relies on fixed, rule-based programming, agentic AI acts independently within set parameters, mimicking human-like decision-making.

Beyond task automation, what complex, multi-step problems are agentic AIs uniquely positioned to solve that current AI models cannot?
Agentic AI is equipped to handle complex, multi-step problems that require autonomous reasoning, adaptability, and real-time decision-making. Agentic AI utilises agents that are capable of driving efficiency, security, and automation in IT operations.

• Proactive problem resolution: Traditional IT monitoring tools are reactive, meaning they only alert IT administrators after an issue occurs. Agentic AI, however, can predict system failures based on trend and automatically implement preventive measures before failures impact operations.
• Enhanced security and compliance: Cyberthreats evolve constantly, and static security policies are insufficient. Agentic AI enhances security postures by continuously analysing network behaviour, detecting anomalies, and implementing corrective actions in real time.
• Optimised IT resource allocation: By analysing workload trends and infrastructure utilisation, agentic AI dynamically reallocates IT resources to ensure optimal performance and cost efficiency.
• Real-time anomaly detection: Network performance issues and fraudulent activities often go unnoticed until they cause significant damage. Agentic AI detects anomalies in real time and mitigates potential risks before they escalate.
• Seamless workflow automation: Agentic AI streamlines complex IT workflows by automating repetitive tasks and integrating with IT service management (ITSM) solutions.

How much autonomy should Agentic AI have in decision-making, and where should humans remain in the loop?
In multi-agent systems, agentic AI entities collaborate by specialising in tasks; sharing information; and coordinating using predefined protocols, reasoning models, and shared memory. However, for risky operations such as updating databases or merging code, agentic AI should defer to human approval. To enable this balance, it’s essential to clearly define the automation level allowed for each action, ensuring that agentic AI acts independently where safe and involves human oversight where necessary. This structured autonomy supports effective human-AI collaboration while minimising risk.

What infrastructure challenges (such as power, cooling, latency) arise when deploying Agentic AI at scale?
Deploying agentic AI at scale demands huge infrastructure demands, especially in power, cooling, and computing. Agentic AI requires high computational power due to the dense AI workloads. These systems rely on GPUs that require more energy consumption, thereby pushing data centres to implement thermal management like liquid cooling and redesigning racks to handle increased weight as GPU servers are heavier than standard servers.

Conventional cooling methods are inadequate. Direct-to-chip cooling is accomplished to maintain performance without thermal throttling. The compute density also requires new architectural approaches, such as modular or prefabricated AI-specific pods.

As these agent systems function autonomously and interact in real time, managing latency and orchestration across distributed AI agents is a challenge. They require rapid access to vector databases and APIs, which requires low latency and better network design.

The existing cloud infrastructure was not built for agent workflows, which leads to bottlenecks in both data movement and contextual memory. This leads organisations to invest in faster networking, hybrid memory layers, and specialised agent run times to ensure scalability.

What’s the next frontier for Agentic AI—will we see AI “agents” collaborating like human teams?
The agentic era is indeed promising. This evolution will be particularly transformative in work-intensive fields like IT support, customer service, and software development, where Agentic AI can take on repetitive, time-consuming tasks. Rather than replacing human roles, Agentic AI will augment human capabilities, fostering a positive relationship between technology and the workforce, driving productivity.

Will open-source models keep pace with proprietary Agentic AI systems, or will there be a widening gap?
Even though proprietary agentic AI systems hold power in enterprise support, the open source model is rapidly closing this gap especially in agentic AI frameworks and fine-tuning capabilities. As they are driven by community-led innovation, more customisation, cost-efficient, transparent, and ethical AI.

It’s not a clear “no” that open-source won’t keep pace; rather, it’s a dynamic landscape where both approaches will push each other forward, leading to a more diverse and capable AI ecosystem. The gap might not widen significantly, and in some specialised areas, open-source could even lead.