As AI technology rapidly advances, it’s no longer just about accelerated adoption but also the transformation of its possible functions. We’re observing a tectonic shift from automation to autonomy.

Take, for instance, the groundbreaking experiment, GPT-3.5, when wrapped in an agentic loop, increased the coding accuracy from 48.1% to 95.1%. Conventional AI systems (gen AI included) have evolved to a revolutionary frontier, agentic AI – transcending simple prompt-response interactions and achieving autonomy and cognition reminiscent of the human intellect. This advancement signifies the capability to independently reason, initiate actions, and produce high-quality results with minimal human oversight, bringing us closer to creating the super-intelligent AI— Artificial General Intelligence (AGI). AGI can further redefine automation processes that are already in use, like Application Programming Interfaces (APIs), Robotic Process Automation (RPA), and Intelligent Document Processing (IDP).

Yet, challenges of explainability, reliability, adaptability, and ethicality remain. Amidst this, the difference between the decision-maker and the agent in decision-making is essential, and understanding agentic AI to its core becomes crucial.

Agentic AI: Agents of autonomy

The core of agentic AI is different from traditional AI, which is largely (if not entirely) dependent on algorithms. Beyond training, agentic AI can adapt to its environments and execute complex decision-making tasks. As a result, traditional AI is now being considered Narrow AI because agentic AI is miles ahead in terms of context understanding, human-like interactions, complex analytical skills, and more. To illustrate,

In insurance, agents can assist in diverse tasks, from balancing the needs of all stakeholders for quick insurance claims to predicting customer queries based on previous interactions, making claim processing faster and more transparent, reducing waiting times, maintaining consistency in customer service, and more.

In healthcare, agentic AI can dynamically identify several effective drug combinations and even predict patient responses based on their genetic history and medical condition. This makes collaboration more precise and personalized without relying on guesswork and prolonged discussions.

In software engineering, by assigning agents to different roles, such as that of a business analyst, developer, tester, etc., the agents can improve the quality of the code generated, eliminate human error, reduce decision-making time, and ensure that it meets the requirements.

Foundations for a sophisticated agentic AI

While the opportunities are ample, there are also several degrees at which agentic AI operates.

• The complexity of the task: AI agents are designed to tackle complex concerns by synthesizing information from various sources (physical and digital), making informed decisions, and providing comprehensive solutions, proving to be more versatile and valuable.
• Diverse capabilities: AI agents offer a broad range of functionalities across different domains. This makes them highly beneficial in environments that require the ability to switch between tasks seamlessly.
• Adaptability to the new: One significant advantage of agentic AI is its ability to learn and adapt to new and unforeseen situations. This ensures that the agent can respond to dynamic and evolving requirements, continually improving its performance when presented with new data and experiences.
• Independent execution: Agentic AI systems operate autonomously, executing tasks and making decisions without the need for constant human oversight. This independence is crucial in scenarios where timely and accurate actions are necessary, like in the case of autonomous vehicles.

Whatever the capability, the foundation of agentic AI is built on a few core components: the agents and the multiple environments they interact with. These can be physical, virtual, or both. The third important aspect of this foundation is the shared memory, which allows total coordination amongst agents (there can be multiple agents within a single ecosystem). This is the hotspot where information, strategies, and goals are exchanged, and every agent contributes and benefits from this shared memory.

Today, many frameworks help develop agentic AI applications.

• AutoGen: Developed by Microsoft, it provides a multi-agent conversation framework where these agents can converse with each other to solve tasks.
• Maestro: A framework designed to coordinate the likes of Claude Opus, GPT, and local LLMs to be cohesive subagents.
• Crew AI: Built on LangChain, it helps create and manage collaborative and specialized AI agents.
• TaskWeaver: Built by Microsoft, it’s a code-first agent framework specialized in data analytics tasks.
• LangGraph: A LangChain offering, it helps build multi-actor applications in a graph network, ensuring seamless integration and collaboration.

At HTC, our in-house industry-agnostic MAGE Platform, with robust AI capabilities, leverages these frameworks to develop agentic AI solutions across verticals like insurance, healthcare, retail, and more.

Autonomy paves the way for transformation

We are currently at a stage, where research suggests that 60-70% of the work hours in our global economy could be automated, and agentic AI can take it a step further. The benefits are multifold: higher-quality outcomes, efficient time usage, and even improved user preference solicitation like a personal assistant AI sending clarifying doubts in natural language and at strategically appropriate times.
These could easily result in enhanced decision-making, elevated human-AI collaboration, personalized customer experiences, automating operational processes, increasing workforce productivity, and ensuring ethical and sustainable practices.

More so, agentic AI is scalable. A traditional AI may simply help automate processes in claims management. However, an agentic insurance tool can unify underwriting, claims, fraud detection, and other aspects with a multi-agent system, giving way to AI transformation for the organization.

The ethical dilemma

Yet, autonomous decision-making raises multiple questions. How do we ensure mutual respect between humans and advanced assistants over user autonomy and well-being? In a world that’s becoming increasingly automated, does the thin line between decision-making agents and decision-makers eventually blur? Concerns of legality, accountability, liability, fairness, transparency, privacy, and ethical decision-making come into play.

Without careful human intervention, the entire system can potentially collapse, if not result in a catastrophe (or more). At HTC, we continuously strengthen our ethical and secure AI capabilities with data encryption, threat detection, human autonomy, decision supremacy, and more.

What does the future look like?

Agentic AI is transforming daily, and agents are becoming even more prevalent. Human capabilities are also augmenting at the same rate, if not faster. At the cusp of this, GAINs (GenAI Networks) emerged to enable collaborative problem-solving.

Our future is about to look somewhat like this: multi-agents work 24/7, employees balance work and life, and leaders focus solely on strategizing and scaling. At the same time, agentic AI systems function with Human-in-the-Loop (HITL) approaches, ethical and legal governance frameworks, and training and sensitization programs that oversee these AI systems.

Now, coming to the million-dollar question: Businesses are set to transform and evolve with automation. But does decision-making become easier? Probably not. That’s what makes us human. But are agents essential in making us resilient? Definitely.