Ashraf El Zarka, the Vice President and Regional Managing Director of MEA & Pakistan at UiPath, says human-in-the-loop design isn’t optional—it’s foundational to building trust, ensuring accountability, and scaling agentic AI responsibly
How do you define Agentic AI, and how does it fundamentally differ from traditional chatbots or rule-based automation?
Agentic AI departs from deterministic automation by introducing autonomous, goal-oriented agents capable of reasoning, planning, and adapting in real time. Traditional chatbots operate on scripts or retrieval, constrained by fixed flows and brittle logic. Agentic AI interprets intent, maintains memory across complex dialogues, and executes decisions across systems without human prompts.
UiPath empowers these agents by embedding them into orchestrated workflows that integrate AI agents, robots, and people into an intelligent system. This ensures every task is contextually grounded, policy-aligned, and outcome-driven. The difference isn’t incremental—it’s foundational. We’re moving from tools that function in a rules-based, deterministic paradigm to one that is probabilistic and powered by controlled agency.
This paradigm shift is exemplified by the UiPath partnership with Google Cloud, where a generative AI-based Medical Record Summarization agent was developed. Powered by Vertex AI and Gemini models, this agent significantly reduces prior authorization turnaround times by up to 50%, showcasing the real-world impact of agentic AI in complex, regulated environments.
Beyond task automation, what complex, multi-step problems are agentic AIs uniquely positioned to solve that current AI models cannot?
Agentic AI unlocks what conventional AI can’t: the autonomy to execute end-to-end, dynamic workflows. Think of document-heavy onboarding in regulated industries, fraud detection across siloed systems, or claims adjudication that hinges on policy nuance, historical context, and external constraints.
At UiPath, we’ve seen AI agents take on specialized roles within workflows—retrieving data, interpreting context, applying logic— before feeding insights into broader decision making. This modular, orchestration-first approach delivers continuity, speed, and resilience, supporting teams in solving problems faster and more effectively than other types of AI tools allow.
How much autonomy should Agentic AI have in decision-making, and where should humans remain in the loop?
Autonomy must be earned, not granted. At UiPath, we believe in controlled agency—an architecture where agents operate within clear boundaries and humans remain in strategic control. Tasks with high predictability can be delegated fully; complex or regulated actions demand human validation. Our agentic orchestration layer, UiPath Maestro, enforces this dynamic: agents execute tasks but exceptions, ethical and edge cases route to people.
Human-in-the-loop design isn’t optional—it’s foundational to building trust, ensuring accountability, and scaling agentic AI responsibly. It ensures trust, mitigates risk, and allows autonomy to scale responsibly, without compromising oversight. Robots play a central role in delivering this controlled agency. They serve as secure, deterministic gateways, retrieving only the data agents need, when they need it, and nothing more.
This creates a tightly governed environment where automation is both powerful and predictably safe.This approach is evident in the UiPath collaboration with HCLTech, aiming to accelerate agentic automation across industries. The partnership focuses on deploying pre-configured AI agents with built-in controls, ensuring seamless scalability while maintaining necessary human oversight.
What infrastructure challenges (such as power, cooling, latency) arise when deploying Agentic AI at scale?
Agentic AI is compute-hungry and coordination-heavy. Unlike single-model deployments, multi-agent systems introduce orchestration latency, memory state demands, and model-inference complexity across nodes. Power and cooling footprints rise, particularly when agents rely on on-premise GPUs or real-time edge execution.
At UiPath, we mitigate this by leveraging hybrid architectures—balancing cloud scalability with on-device precision. Moreover, automation robots serve as secure data intermediaries, reducing bandwidth loads by delivering only the data AI agents need, when they need it. Infrastructure must evolve not just to support AI, but to sustain intelligent automation ecosystems.
The integration of UiPath’s platform with Microsoft Copilot Studio exemplifies this evolution. The bi-directional integration facilitates seamless interaction between AI agents and automations, enabling customers to automate complex end-to-end processes and improve scalability, thereby addressing infrastructure challenges inherent in agentic AI deployments.
What’s the next frontier for Agentic AI—will we see AI “agents” collaborating like human teams?
We’re already there. Agentic systems are no longer single-threaded tools—they are emerging as collaborative networks. One AI agent gathers data, another interprets, and a third acts. Together, they augment human workflows by handling repetitive, time-consuming steps—freeing people to focus on judgment, strategy, and creative problem-solving.
UiPath enables this through agentic orchestration—a paradigm where agents, robots, and humans form hybrid teams. The frontier isn’t just smarter automation—it’s autonomous collaboration. Over time, these agents won’t just assist humans—they’ll assist each other, continuously improving and co-evolving. This unlocks enterprise resilience at a scale we’ve never seen before.
Will open-source models keep pace with proprietary Agentic AI systems, or will there be a widening gap?
The gap won’t be in capability—it will be in control. Open-source models are rapidly catching up in sophistication. But enterprise readiness demands more than performance—it demands governance, security, interoperability. UiPath integrates commercial and open ecosystems—but surrounds them with guardrails: access controls, audit trails, and escalation paths.
Proprietary platforms may advance faster in vertical depth, but open ecosystems offer modular agility. The real differentiator isn’t model architecture—it’s the orchestration framework that binds agents, tools, and humans into one reliable system. That’s where the gap will be felt.
