Agents in the era of AI Plateau
Panos
Co-founder & CEO
December 6, 2024
in
🏆 Leadership Insights
The landscape of artificial intelligence (AI) is rapidly evolving, but are we reaching a point where the progress is leveling off? AI Agents are a breakthrough that goes beyond traditional LLMs and transforms how businesses interact with customers. Let's explore what AI Agents are, how they work, and why they matter now more than ever.
What is an AI Agent?
Think of an AI Agent as an advanced application that combines multiple LLMs, workflows, and tools to achieve specific goals autonomously. It breaks down complex problems into smaller tasks, plans solutions, picks the best tools for each step, takes action, and finally integrates everything into a seamless, meaningful response.
In essence, it’s like having a tireless SDR, salesperson, and customer support representative rolled into one—always on, always ready to tackle complex challenges and deliver tailored, impactful solutions.
How do AI Agents work?
AI Agents are built on large language models (LLMs), but they go several steps further. Traditional LLMs rely solely on pre-trained data, often limited in reasoning and flexibility. In contrast, cutting-edge AI Agents, like those powered by Moveo's Agentic technology, dynamically call on backend systems to fetch the latest information, streamline workflows, and autonomously solve complex subtasks to accomplish an ultimate goal – given by you!
Here’s where it gets interesting: These agents learn to adapt over time. By remembering past interactions and anticipating future needs, they deliver experiences that feel increasingly tailored and personal—setting a new standard in customer engagement.
AI Agents in Action: A BFSI Use Case
Imagine a customer typing, "I want to cancel my credit card," alongside an image of their latest credit card statement with charges highlighted. A basic chatbot might provide step-by-step instructions for cancellation—straightforward, but transactional and limited in scope.
An AI Agent takes an entirely different approach. Acting like a skilled advisor, it initiates a thoughtful dialogue to uncover the reason behind the request. Are the charges incorrect? Is the customer dissatisfied with the benefits? By engaging in a meaningful conversation, the Agent can propose personalized solutions, such as waiving fees or highlighting unused rewards, to retain the customer.
Behind the Scenes: How AI Agents Solve Problems
What makes AI Agents so effective? To illustrate further how the AI Agent becomes far more effective than a simple chatbot, it’s worth examining the steps it would take to address the above seemingly simple question. Unlike a chatbot that delivers a static response, the AI Agent leverages multiple LLMs working in concert, each specialized for a specific task, to provide a nuanced and impactful solution.
It’s all about their ability to orchestrate specialized tools and processes. Let’s break it down:
The process might begin with a planning LLM that deconstructs the customer request to determine which tools to use:
Tool 1: Analyze the customer’s uploaded image with highlighted charges to determine which fees might have triggered dissatisfaction
Tool 2: Search an internal knowledge base to provide relevant information on the customer's credit card and how it works
Tool 3: Call an API to analyze the customer’s habits and understand their specific usage patterns
Tool 4: Use a calculator to illustrate the benefits or savings of keeping the card
Once these tasks are completed, a response-generation LLM synthesizes the results into a human-like and empathetic response.
For instance, if the tools reveal that the user has highlighted a late fee, is a loyal customer, and has significant unused rewards points, the AI Agent might respond with:
"I completely understand how late fees can be frustrating. Since this is the first time this has happened and you've been such a loyal customer, I’d be happy to waive that fee for you. To make things even easier moving forward, I’d recommend setting up autopay to avoid similar situations in the future. Also, I noticed that you’ve already accumulated 80,000 reward points using your card—these can be redeemed for up to $800 in flight credits!"
In real-world use cases, these sequences involve many more "steps" and additional specialized LLMs, each performing distinct tasks. GenAI agents built for enterprise applications are particularly complex, as they require a sophisticated orchestration layer to operate seamlessly in sub-second response times, improve continuously through human feedback, and uphold strict data privacy standards.
Why Not Wait for GPT-5 to Handle It All?
It’s tempting to assume that future LLMs like GPT-5 will handle everything seamlessly. But here’s the challenge: progress in LLM performance is slowing. Benchmarks like Massive Multitask Language Understanding (MMLU) show only marginal gains since GPT-4’s release—GPT-4 achieved 86.4%, and OpenAI's latest reasoning model, GPT-o1-preview, reaches about 90%, a modest improvement.
A key issue is diminishing returns on additional training data, as highlighted in No “Zero-Shot” Without Exponential Data. Simply put, the amount of high-quality public text available is finite, and newer models require exponentially more data to deliver significant improvements.
While GPT-5 might offer some advancements, it’s unlikely to single-handedly redefine the AI landscape. Real innovation will come from breakthroughs in architectures and multimodal learning—not just scaling up existing models.
Moveo’s Approach: AI Agents represent the next revolution in Customer Engagement
At Moveo.AI, we believe the future lies in smaller vertical-specific models!
AI Agents aren’t simple LLMs; they represent the most reliable path to achieving significant and quantifiable business outcomes using Generative AI.
Our understanding of the technology and the evidence at hand have led us to establish that the winning strategy will not come from relying on a single LLM or incremental improvements to it. Instead, success lies in developing applications that orchestrate complex, vertical-specific LLMs—what we call AI Agents. As a company, we aim to transform how businesses communicate with their customers by turning one-way notifications into conversations with proactive GenAI Agents.
By utilizing an ensemble of LLMs and orchestrator techniques, we are able to create AI Agents that beat GPT-4 on several key CX metrics and deliver tangible outcomes to more than 100 customers.
Traditional, one-way notifications—like payment reminders or generic promotions—often fail to deliver on their potential. They’re static, generic, and lack the context to truly engage customers. Our AI Agents turn these moments into dynamic, hyper-personalized, and context-driven conversations.
By leveraging GenAI, we create Agents who don’t just deliver messages—they anticipate customer needs, engage them with relevant information, and, most importantly, drive business outcomes like increased revenue, higher adoption rates, and better retention.
⚠️Try Moveo's agent for BFSIs here
Our proactive approach enables businesses to move beyond communication as a task and turn it into a strategic tool that creates value at every touchpoint and throughout the customer lifecycle: from acquisition all the way through to debt collection.
Why Focus on BFSIs?
The BFSI sector presents unique opportunities for innovation. Moveo’s AI Agents are purpose-built around 3 core use cases:
Product Activation and Education
Churn Reduction
Humanized Debt Collection
With the right tools—like sales scripts, objection handling, friendly reminders, and access to user profile information and recent transactions—Moveo’s proactive and goal-oriented Agents empower BFSIs to facilitate engaging, hyper-personalized conversations that not only inform users but also encourage them to take specific actions, unlocking new revenue streams for enterprises.
Conclusion
AI Agents are more than just a technological advance—they act as catalysts, enabling Generative AI to create transformative shifts similar to those sparked by the adoption of cloud computing and mobile technology.
Our goal to turn one-way notifications into conversations with proactive GenAI Agents for BFSIs is only getting started!