🦸🏻#5: Building Blocks of Agentic Systems

Intro

We took an unconventional approach to launching our Agentic Workflow series, beginning with open-endedness – a concept that lays the foundation for understanding the dynamic potential of agentic systems. From there, we introduced the essential vocabulary, along with real and potential examples of agents, to build a solid base. Now, it’s time to dive into the building blocks of agentic systems and explore the core components that bring these systems to life. What does it take to "wire together" an agent to make it work?

In today’s episode:

1. Different frameworks for agentic systems

2. Critical components (Profiling, Knowledge, Memory, Reasoning&Planning, Reflections, Actions)

3. Human-AI communication as a trend (review of “Mutual Theory of Mind for Human-AI Communication”)

4. Conclusion

5. Bonus: Resources to diving deeper

In their early days, agents were more like isolated bots, each designed for a narrow set of tasks. Fast forward to today, and it’s all about creating interconnected autonomous systems that fully leverage the ever-growing capabilities of AI.

As a quick reminder, autonomous agents are entities that perceive their environment, make decisions, and act to achieve goals. These agents vary in their learning capability, physicality, specialization, and task complexity.

When it comes to describing the core components of these systems, there are multiple approaches. For instance, some frameworks are more detailed, like this one:

Others, like LangChain’s Harrison Chase, take a more schematic and simplified approach:

Regardless of the framework, successful implementation of AI agents boils down to a few critical components →