Following is my definition of what an ‘Agentic AI’ should be,
- User Persona Adaptation: The agent should recognise the specific user by their persona in context of the application, style of conversation, past conversational output and its inferences, user activity data and inferences derived from them.
- Reflective Response Generation: The agent should be able to replicate the ‘human thinking’ effect with help of a software framework leading to reasoned and specific answers.
- Comprehensive Memory System: The underlying LLM should be enhancing the agent to output answers that are processed with infinite context derived from past conversational data, web search, vector databases, and user activity.
- Tool Utilisation Capability: The agent should be able to perform actions with the help of APIs, databases, analytics tools, mathematical tools, media communication tools, localisation tools and many other tools serviced internally or by 3rd parties.
- Human-in-the-Loop Integration: The User Interface and Experience should allow easy plug-and-play of user, human expert and agent ternary system. Humans have to perform the task of oversight, value-addition, and maneuvering to ensure the best experience for the user.
- Trust Building Mechanisms: The agent and the human expert should be able to work together to win trust of the end-user. This has to be accomplished by citation, stepwise response and such technological solutions.
- Human-like Interactions: The human-agent system should be able to enhance the conversation with end user such that it replicates the empathy, emotion, and charisma that reflects out of a human-human conversations. This should ultimately define the Aha Moments within any Agentic AI use case.