Designing Agentic AI Systems: From Prompting to Autonomous Workflows

AI is evolving beyond assistive copilots toward autonomous, agent-driven systems capable of handling complex, goal-oriented tasks. While copilots enhance productivity, they often fall short in scenarios requiring reactive decision-making and independent execution .

In this session, we will explore why building your own agents is becoming essential and how to move from basic prompting to structured, goal-driven workflows . We’ll cover key concepts such as advanced prompting strategies, incorporating memory, and agent-driven workflow orchestration , enabling systems to dynamically manage and execute tasks.

Designed for a broad audience, this session provides both conceptual clarity and practical insights into building real-world agentic AI systems.

Key Takeaways:

• Limitations of Copilot AI – understanding why autonomous agents are needed
• Advanced Prompting – moving from simple prompts to structured reasoning workflows
• Memory in Agents – enabling context retention and informed decision-making
• Agentic Workflows – designing dynamic, agent-driven execution systems
• System Design Perspective – building scalable and practical agentic AI solutions

Sarvagya Agrawal is a Software Engineer at American Express, working at the intersection of AI, NLP, and agentic systems.

He has led multiple AI initiatives across anomaly detection, accessibility, and intelligent automation, and actively contributes to the AI community through technical writing and global talks. His interests include building next-generation AI agents, scalable ML systems, and decision-centric AI architectures.