Building agents that think like humans

Building AI agents isn’t about better prompts - it’s about better thinking. This talk shows how to design agents that reason, plan, remember, and reflect like humans, turning LLMs from reactive chatbots into autonomous, decision-making systems.

Today’s AI can talk-but it doesn’t think. This talk talks about next leap in AI: agents that reason, plan, remember, and reflect like humans.

You’ll see how to move beyond prompt-driven chatbots and design agentic systems with goals, memory, decision loops, and self-critique - systems that pause, think, and adapt before acting. Through concrete architectures and real-world examples, we’ll break down how human cognition inspires better AI behavior, fewer failures, and far more useful outcomes.

This talk is designed to be highly practical and architecture-focused, not theoretical. It draws from my hands-on experience building production-grade LLM systems including agentic workflows, RAG pipelines, memory-driven agents, and tool-augmented reasoning systems. I’ve worked extensively on designing decision loops, reflection mechanisms, and failure-handling strategies for real-world use cases, not just demos.

The session does not require any special technical setup beyond standard presentation support. No live internet access is mandatory, though a short optional demo or architecture walkthrough can be included if preferred by the event. Slides will focus on clear diagrams, mental models, and implementation patterns that attendees can directly apply.

I am well-suited to speak on this topic because my work intersects at the intersection of engineering and system design - translating abstract ideas like “human-like thinking” into concrete, buildable agent architectures. I regularly explain complex AI concepts to mixed audiences (engineers, product leaders, founders), and I prioritize clarity, realism, and honesty about limitations and trade-offs.

This talk is intentionally scoped to avoid hype and instead equip the audience with a grounded understanding of what actually works when building agents today, why many agent implementations fail, and how to design systems that behave more predictably and responsibly.