From Hype to Strong Foundations: What the Rise, Fall and Resurgence of Agents Can Teach Us About Outlasting the Cycle

The presentation titled "From Hype to Strong Foundations: What the Rise, Fall and Resurgence of Agents Can Teach Us About Outlasting the Cycle," was delivered by Aditya Kumarakrishnan, a Technical Fellow at Walmart Global Tech. Aditya shares insights from over two years of experience in building agents, highlighting key learnings and future directions in the development of agent systems.

Main Themes Discussed:

• Inevitability of Agents: Despite the recurring cycles of hype and disillusionment, the conceptual foundation of agents remains a fundamental part of computing, stretching back to the ideas of early AI pioneers like Alan Turing and John McCarthy.
• Four Foundational Ideas: Aditya emphasizes four critical areas for building robust agent systems:
  1. Embracing a stronger definition of agents.
  2. Creating modular and extensible agent architectures.
  3. Leveraging process science to integrate and execute business processes.
  4. Terraforming environments to be conducive for agents to operate effectively.

Challenges and Solutions:

• Environment Suitability: Emphasizing the need to reshape the computing environment around agents to ensure their success by implementing guardrails and providing a more ergonomic setup.
• Process Science Integration: Utilizing process science to understand and automate complex processes, enabling agents to perform meaningful tasks within enterprises.
• Historical Lessons: Highlighting the importance of learning from past cycles to avoid repetitive mistakes and to build on a solid foundation of engineering principles.

Aditya concludes by reinforcing that while these foundational ideas may seem futuristic, they are rooted in tried-and-tested frameworks that can ensure the longevity and success of agent technologies.

This is the end of the AI-generated content.

2025 has been declared "the year of agents", but so were 1994, 1998, and 2016. In fact, the history of agents is a story of recurring peaks and valleys. We'll situate agents and multi-agent systems as a software paradigm alongside Objects, Actors and Microservices, examine the role LLMs and compound AI systems can play, and build a taxonomy of the various protocols and frameworks around and identify gaps enterprises need to navigate.  The core thesis: agents aren't magic, they're software. And like all good software, they need a foundation of solid engineering principles of abstraction and reliability because those who don't learn from the past are doomed to be swallowed by "The Bitter Lesson" or to re-implement JADE.