As the boundaries of industrial automation expand, robots are beginning to think, adapt, and collaborate, and no longer limited to repetitive tasks. Manish Jha, Chief Technology Officer at Addverb, shares how Agentic AI is transforming robotics into autonomous systems capable of decision-making and real-time problem-solving. In this conversation with Tech Achieve Media, he explores how these intelligent systems are redefining human-robot interaction, enabling flexible automation across sectors, and tackling the challenges of real-world unpredictability with a layered, context-driven approach.
TAM: How do you see Agentic AI redefining the role of robots in dynamic industrial environments from task executors to autonomous decision-makers?
Manish Jha: Agentic AI marks a shift from robots simply following instructions to acting as autonomous collaborators within industrial ecosystems. These systems are observing, reasoning, and adapting in real time rather than just carrying out pre-coded tasks. Agentic artificial intelligence can now observe its surroundings, analyze real-time data, and autonomously modify its activities. These systems orchestrate multistep processes end-to-end, learning from every result, employing the fusion of large language models with intelligent agent architectures. The result is resilience and autonomy. Agentic AI is not just enhancing performance; it is redefining the very nature of industrial autonomy, transforming static operations into responsive, intelligent systems.
TAM: How does Agentic AI elevate the human-robot collaboration beyond safety and efficiency?
Manish Jha: Agentic AI has significantly improved human-robot collaboration by enabling robots to act as proactive, autonomous partners rather than passive tools. Instead of rigid, rules-based programming, robots now interact like adaptive teammates. They can take initiative, anticipate human needs, and collaborate across evolving workflows. This enables far more than process efficiency; it unlocks shared decision-making. Most RPA implementations max out at 30-40% task coverage due to their limitations. Agentic AI extends that frontier. It enables robots to work in layered automation, whether sequential, parallel, or iterative, tailored to unique operational demands. This allows them to work alongside human workers and other robots more seamlessly, reducing downtime and operational bottlenecks. For industries dealing with volatility or scale, this level of responsive partnership is a game-changer.
TAM: What are the core technological components that enable Agentic AI to mimic human-like reasoning? How has Addverb integrated these into your robotics systems?
Manish Jha: Agentic AI uses perception, memory, decision-making, and continuous learning along with cognitive technologies to emulate human-like thinking. It draws on contextual understanding based on sensor data, adaptive planning algorithms, and feedback-driven improvement cycles.
At Addverb, we’re integrating Agentic AI into our robotics platforms to enable proactive responses rather than reactive operations. In a dynamic warehouse, for example, robots like AMRs and Robotic Sorters can now autonomously plan optimal routes, adapt to changing warehouse conditions, as well as effectively collaborate with other robots or human workers. Agentic AI is allowing robots to evolve into intelligent agents capable of dynamic decision making.
TAM: A recent Gartner study predicted that over 40% of Agentic AI projects are predicted to be canceled by the end of 2027. What are your views on this?
Manish Jha: The initial hype cycle of any revolutionary technology is usually associated with unrealistic expectations, and agentic AI is not an exception. The projects that are not able to describe the issues that they are trying to address or that have not fully appreciated the complexity of real-world implementation often face delays.
The recent Gartner survey highlights that the cancellations demonstrate the gaps in implementation strategy. At Addverb, we have succeeded in domain-specific integration, constant simulation, and gradual implementation. Instead of approaching Agentic AI as a one-size-fits-all solution, we integrate it into workflows that are created with a deep understanding of operational issues. This quantitative approach has produced strong momentum in applications like autonomous navigation, adaptive task scheduling, and predictive maintenance. We believe that the key requirements to achieve long-term success lie in the combination of technical innovation with context-specific implementation and the development of robust human-in-the-loop systems.
TAM: Agentic systems must function in unpredictability. How do you ensure these machines continue to make safe, optimized decisions even when faced with incomplete data or novel situations?
Manish Jha: At Addverb, safety is designed into every level of our systems. When data is incomplete or unexpected events occur, robots don’t make blind decisions; they follow a structured fallback process. The system assesses confidence levels in real-time. If uncertainty is high, it automatically slows down, reroutes, or hands control back to human operators. We also train our AI using simulated edge cases, so robots are prepared for rare but critical scenarios before they ever hit the floor. It’s a layered, tested approach that ensures machines stay reliable, even when the environment isn’t.
TAM: How does Addverb customize its Agentic AI capabilities to adapt to diverse use cases across industries?
Manish Jha: Each industry operates with its own workflows, safety, and operational needs. At Addverb, we have configured our Agentic AI architecture to be modular and scalable, hence enabling smooth customization. Using AI agents trained on context-specific data, we can make our robots fit the industry-specific standards.
Our approach combines domain-specific trained models, task hierarchies that can be customized by the customer, and user interfaces that are easy to use, giving customers the capability to tune performance without having to code. We also facilitate cloud fleet learning, whereby what is learned in one site can be used to make improvements in all the others. The method will help keep the core AI engine the same, but customize the application of the engine. Through grounding our AI in operational realities, we deliver value that’s not just intelligent but contextually relevant and future-ready.
