In the book Agentic Artificial Intelligence Harnessing AI Agents to Reinvent Business, Work, and Life by Pascal Bornet et al introduces us to the three keystones of Agentic AI: actions, reasoning and memory. It was published in 2025. It is Copyright 2025 by Pascal Bornet, Jochen Wirtz, Thomas H. Davenort, David De Cremer, Brian Evergreen, Phil Fersht, Rakesh Gohel, and Shail Khiyara. It’s a practical, non-technical guide for business leaders, professionals, and curious minds. It’s not a programming book. In the book there is a chapter dedicated to each of these three keystones.

AI agents have impressive capabilities and limitations. They are really “smart” but can’t learn from experience, take actions without thinking them through, and lack basic common sense. They do things that look logical in isolation, but make no sense in context.

These three keystones transform AI agents from just tools to teammates. The future is about AI agents that act, think and learn (and improve) alongside us humans. The term “keystone” is used here to highlight Action, Reasoning, and Memory as essential structural elements of agentic AI — the foundational parts that hold the system together and make it functional. These three are crucial elements that give structure and coherence.

The Three Technical Pillars (Common in AI Frameworks)

In AI development, particularly in frameworks like LangChain or OpenAI’s Agent Mode, agentic systems are often broken down into three functional components:

• Reasoning: The AI’s ability to plan, make decisions, and choose the next step toward a goal.
• Actions: The execution of specific steps, which might involve calling APIs, searching databases, or triggering processes.
• Memory: Retaining information from previous steps or sessions to build context and improve future performance.

These are the building blocks that give an AI the operational capacity to deliver on its goals. Without them, an agent can’t move beyond simple, one-off interactions.

How They Work Together

While the two frameworks use different language, they’re complementary. The technical pillars (Reasoning, Actions, Memory) provide the mechanics that make the conceptual keystones (Autonomy, Proactivity, Adaptability) possible. For example, an AI’s adaptability depends on its memory, and its proactivity often relies on strong reasoning and the ability to take action autonomously.

Coming Next

In the next post, I’ll explore practical applications of these ideas, showing how agentic AI could be applied in projects ranging from sustainable development initiatives to education and local community problem-solving.