The Gist
- Dynamic AI ecosystems. Software applications are shifting from static, monolithic software to dynamic, AI-powered ecosystems that integrate seamlessly with diverse business needs.
- Agentic AI disruption. Agentic AI systems may begin to neutralize the growth of SaaS applications as enterprises look to leverage AI to further optimize their business
There is growing momentum around an artificial intelligence (AI) paradigm called agentic AI, which has the potential to transform businesses by reducing workloads, simplifying processes and orchestrating work. Agentic AI is a type of AI that can act independently and make decisions without human intervention. Agentic AI systems can analyze data, predict outcomes and execute decisions autonomously. Think of AI agents as digital teammates that people can collaborate with using natural language or prompts.
Like most AI paradigms, agentic AI is not an entirely new concept. It has roots in the 1950s and 1960s when AI efforts focused on creating programs that could mimic human decision-making. In the 1980s and 1990s, advancements in robotics and computer vision gave machines agent-like qualities.
Early examples of agentic AI were also found in AI chatbots that demonstrated the potential for AI to interact with customers autonomously. These early chatbots were limited to using pattern matching and decision trees to simulate conversation.
In 2024, the concept was reinvigorated by technology entrepreneur Andrew Ng. Ng has spent his career in AI, and he’s been the co-founder of Google Brain, chief scientist for Baidu and an adjunct professor at Stanford. He’s helped launch some of the most ambitious AI programs and has secured his status as one of the most respected figures in computer science.
Ng predicts that AI-powered agents will be a big focus in the tech space in the coming years, contributing more to AI progress than simply scaling up large language models. He’s observed that more tech companies are developing platforms to support agentic workflows, with a growing number of applications being built around them. Multi-agent systems can solve more complicated problems than traditional single-agent setups.
Table of Contents
Are LLMs Similar to AI Agents?
Today, we use AI to generate sophisticated responses to prompts providing key insights in seconds. What used to take hours of research and synthesis can literally happen in real time with thoughtful prompting. However, these responses are linear in nature, or “zero-shot” in AI terms, where the answer is provided from start to finish with no asynchronous or iterative actions.
LLMs can’t make their own decisions and can’t plan or adapt to changing situations. Instead of simply generating content, AI agents can interact with their surroundings, respond to changes and complete tasks with minimal human guidance. LLMs don’t adapt or learn in real time; they rely mostly on the patterns they learned during training. LLMs can’t interpret a physical environment and only react to input given, not the context around them.
AI Agents: Set it and Forget it
AI agents can act independently without constant input from humans. AI agents can execute a “chain of thought” that requires complex orchestration and iteration among different systems and processes. This means that an agent can deconstruct a complex workflow into a series of intermediate steps, facilitate additional analysis and then synthesize it into a coherent answer or outcome. Think of it like a robot that operates without a human controller, determining what to do next based on its surroundings.
AI agents offer capabilities that extend beyond the capabilities of LLMs. AI agents offer functionality that LLMs alone cannot match, as they have connections to various data sources and tools to orchestrate decision-making. For example, an LLM can generate text based on a prompt, whereas an AI agent can take the generated output, analyze additional data and perform a series of sequences of actions to complete a broader objective or task.
AI Agents vs. Chatbots
AI agents are more sophisticated than traditional AI chatbots. Chatbots are primarily designed to answer pre-defined questions or follow a set script, while agentic AI can analyze information, learn from data and make decisions based on context to achieve goals. Traditional AI chatbots are rule-based, which limits their proactiveness and autonomous capabilities. This creates a limited scope of what can be accomplished through these rudimentary AI interfaces.
An AI chatbot works well to provide answers to basic questions on a product or service through a website or mobile app. However, an agentic AI system can monitor market trends, identify potential opportunities and automate product purchase decisions based on pricing and other pre-determined criteria.
Agentic AI is the next evolution of generative AI.
Related Article: The Rise and Impact of Agentic AI: Transforming the Human Experience
Core Components of Agentic AI
Agentic AI Systems Are Task and Goal-Oriented
Once enabled, they can act autonomously and make decisions without human intervention. They are adaptive and can learn from their experiences and actions over time. AI agents can collaborate and work together to complete tasks much like human teams. Agents are typically modular in design so they can be scaled to solve complex problems by distributing specialized tasks to sub-agents and then orchestrating the result.
Agentic AI Systems Combine Large Language Models (LLMs), Tools and Prompts for Complex Task Management
An agentic AI system typically has core components, including agent architecture, environment interfaces, task management, communication protocols and learning mechanisms. These components are becoming readily available with software platforms specifically designed to simplify the creation, deployment and management of AI agents.
Agentic AI Systems Are More Powerful Than Basic LLMs
For example, an agentic AI system could autonomously manage an entire loan process for a customer. It goes beyond basic automated processing based on rules, which is what a typical robotic process automation (RPA) tool provides. Instead, it can analyze the customer’s financial history, predict future cash flow, adjust repayment terms, analyze market interest rates and proactively offer a best-fit loan product to the customer. This all happens in real time without requiring manual intervention.
Agentic AI Systems Adapt in Real-Time
Automating customer service requests becomes more efficient than a traditional AI chatbot as AI agents don’t just respond to customer queries; they also autonomously manage complex interactions and guide customers through various processes. AI agents adapt to the situation in real time. They anticipate the customer’s next steps and offer relevant information before it’s even requested. This reduces friction in customer interactions and creates a smoother, more efficient service experience.
Related Article: From Robo to Relatable: Make AI in Customer Service More Human
How Agentic Frameworks Drive AI Innovation
An agentic AI framework provides developers with pre-built components, data models and tools that provide standard, foundational capabilities to streamline the creation of complex AI agent systems. These frameworks speed up development, promote standardization and enhance scalability and accessibility in AI.
Many of these agentic AI frameworks started to appear at the end of 2023 and throughout 2024. The space is still rapidly evolving, but there are some interesting frameworks that have emerged such as LangGraph, Crew AI, Microsoft AutoGen, FlowiseAI and OpenAI Swarm:
LangGraph: Modular Framework for LLM Agents
LangGraph, from the developers of LangChain, provides a modular framework for building LLM-driven agents capable of complex task workflows and supports prompt chaining, memory management and tool integration. LangChain allows multi-step tasks, data source connections and adaptable agent functions for dynamic applications.
CrewAI: Open Source Framework for Creating AI Agents
CrewAI is another open-source framework, backed by Andrew Ng, that allows the creation of AI teams where each agent has specific roles, tools and goals, working together to accomplish complex tasks. This is like assembling an expert team where each member brings unique skills and expertise, and they can collaborate seamlessly to achieve the task objectives.
Microsoft’s AutoGen: Open Source Supports Cooperation for Multiple AI Agents
Microsoft’s AutoGen is an open-source programming framework for building AI agents and facilitating cooperation among multiple agents to solve tasks. AutoGen aims to provide an easy-to-use and flexible framework for accelerating development and research on agentic AI. AutoGen is based on the actor framework where each agent is an actor.
FlowiseAI: Low-Code Tool Supports LLM Orchestration
FlowiseAI is an open-source, low-code tool for developers to build customized LLM orchestration flow and AI agents. Users can design AI agents through a visual interface without writing large amounts of code, which makes it accessible to non-programmers.
OpenAI Swarm: Lightweight Framework and JSON Structures
OpenAI Swarm is an experimental and lightweight framework designed to simplify the creation of multi-agent workflows. It’s more of a “design pattern” of “handoffs between agents” rather than a full-fledged framework. Each agent in Swarm comes with its own set of instructions, a designated role and a collection of available functions. These functions are transformed into JSON structures, which allows for seamless integration and execution.
Many of these frameworks will have similar capabilities but here are some key considerations when selecting an agentic AI framework.
-
Streaming: Does the platform support real-time token or message streaming?
-
Human in the loop: Can a human be introduced in the decision-chain process?
-
Time travel: Can the platform go back in time to understand where agents went wrong?
-
Memory: Does the platform remember decisions?
-
Low code: Is there the ability to create agents?
-
Language support: What languages do they support?
The best way to get a sense of the power of these frameworks is to start building solutions to understand the capabilities of what is possible today.
How Agentic Design Patterns Enhance AI Intelligence
Agentic AI uses a category of design patterns that focus on how AI agents make decisions and interact with their environment, including planning, reflection and tool usage.
Consider asking an AI agent to plan a trip, where it would need to consider factors like travel dates, destinations, accommodation, transportation and activities. Then it could generate a detailed itinerary with logical steps. To accomplish this, the agent may use several design patterns. Several popular design patterns can be used to do this.
Key Design Patterns in Agentic AI
Reflection is iterative refinement where the prompt response can be used to review or critique the output of the LLM response. This forces the LLM to critically analyze, identify flaws and improve the response. This cycle can repeat until the LLM is satisfied with the output. Reflection improves accuracy and reliability and encourages the LLM to explore different perspectives and generate more nuanced responses.
The tool use pattern is a powerful mechanism for allowing LLMs to interact with external systems, APIs or resources. This pattern extends LLMs which may be limited by outdated pre-trained data, and it allows dynamic integration with external resources. It also lets LLMs launch external tools to complete a specific task.
A reasoning or planning design pattern is when the LLM autonomously decides the sequence of steps required to execute a task. This pattern frequently involves interweaving multiple LLMs to accomplish a task. By determining each step, what model to use and what choices to make, it allows autonomous driving of a complex multi-step agentic workflow.
Lastly, multi-agent collaboration dispatches multiple agents in a shared environment to collaborate or compete to achieve a specific goal, with each agent acting independently to solve complex problems. In some cases, LLMs may be specialized to run certain tasks. An example of this might be autonomous vehicles coordinating movements on a road to orchestrate traffic flow.
Related Article: Agentic AI and Marketing: The Death of the Traditional Funnel?
What Lies Ahead for Business Applications With Agentic AI Dynamic?
What does all this mean for the future of enterprise business applications? Is their future under siege? One thing is for sure: Software applications will look different.
It’s quite possible business applications may begin to collapse in the agentic AI era. If you consider the core foundation of a typical SaaS application, it consists of a data repository with a business logic layer and a process-driven front end.
In the future, applications won’t live in silos like today. Consider your current SaaS application footprint for CRM, HR, Finance and IT. Today, these are likely different SaaS applications from different SaaS providers. A lot of effort goes into integrating and correlating these systems for an enterprise to operate its core business processes. Most core enterprise business processes span multiple SaaS solutions in the enterprise, which increases the complexity and effort needed to integrate.
The Shift from Traditional SaaS to AI-Driven Systems
Traditional SaaS often focuses on static, predefined workflows that require customization to change. Agentic architectures are supplemented by AI models that adapt and learn dynamically.
In an agentic architecture, the logic will move to the AI or decisioning layer. Business logic will get compressed into the agent. The AI agent will now orchestrate the work. AI agents will interact with the data repositories directly and make decisions autonomously with limited human interaction. AI will not discriminate which system it utilizes or traverses to complete a task.
AI Agents as the New Interface for Business Applications
This could make the entire interaction model change from process and form-driven experiences to voice-enabled directives. For instance, launching a sales campaign could become as easy as providing a set of instructions such as “Give me a list of my top 20 customers who have done business with me in the past 90 days and send them a promotional offer for a 20% discount on new purchases.” In the background, the agents would work with the systems of record in CRM, marketing and campaign tools to activate this campaign with almost zero touch.
Initially, traditional SaaS applications don’t go away, but they may become “headless,” which means the interaction model changes as agents become the primary user interface.
However, once the AI tier becomes operational, enterprises can begin to think about rationalizing the back ends to consolidate into a unified enterprise repository. AI agents are agnostic to the back end, so they don’t have any coupling to a specific enterprise application. This creates tremendous freedom in enterprise architectures which are frequently shackled by the legacy silos that exist in today’s applications.
The Future of SaaS: Evolution, Not Extinction
We may not see the “death of SaaS” any time soon, but the evolution of software delivery toward more intelligent, flexible and adaptive systems is here. For traditional SaaS and legacy ERP platforms, this marks a shift from static, monolithic software to dynamic, AI-powered ecosystems that integrate seamlessly with diverse business needs. The future of software applications will be how quickly they adapt to these trends to deliver greater value, efficiency and adaptability.
Core Questions Around Agentic AI
Editor’s note: Here are two important questions to ask about agentic AI.
What is Agentic AI?
Agentic AI refers to a type of artificial intelligence that can act independently, make decisions autonomously and execute tasks without constant human intervention. It can analyze data, predict outcomes and adapt to changing environments, much like a digital teammate that collaborates with humans to achieve complex tasks.
How is Agentic AI different from traditional AI chatbots?
Unlike traditional AI chatbots, which are rule-based and respond to predefined queries, agentic AI can interact autonomously with its environment, learn from experiences and make decisions on its own. It is capable of managing more complex tasks, adapting to changes and executing workflows that involve multiple systems and processes, while chatbots are limited to simple, scripted conversations.
Learn how you can join our contributor community.
