Businesses must get ready to work with AI agents if they want to stay competitive.
Many have already adopted them, while others are gearing up to do so. These agents will soon be part of almost every organisation, making up a large global digital workforce.
An AI agent is a software application that engages with its surroundings, collects information, and utilises that data to accomplish predefined objectives.
AI agents are software programs that perform tasks autonomously or semi-autonomously. They run independently to design, execute, and optimise workflows. Guardrails can be built into AI agents to help ensure they execute tasks correctly. They perform tasks with high precision and consistency, and monitor and analyse security threats in real-time, providing proactive measures to prevent breaches and ensure data protection.
According to a Gartner report, AI agents will streamline operations, reduce manual tasks, and improve decision-making processes. It states that by 2027, nearly 15% of new applications will be automatically generated by AI without a human in the loop. Also, AI agents will lead to a flattening of organisational structures, with up to 20% of organisations eliminating middle management positions by 2026.
A Forrester report titled ‘The State of AI Agents, 2024’ states that AI agents are advancing from decision-making to action. It emphasises that businesses will lead the adoption of AI agents, but tech and risks will temper timelines.
According to Forbes, by 2028, agentic AI will be responsible for making 15% of everyday work decisions.
| “We are going to live in a world where there are going to be hundreds of millions of different AI agents, eventually probably more AI agents than there are people in the world.”—Mark Zuckerberg “SaaS apps are nothing more than a CURD database with some business logic, but once the business logic moves to AI agents, SaaS is over.”—Satya Nadella |
Using AI agents for businesses: Important questions
Driving business outcomes with AI agents requires a strategy along with collaboration from enterprise teams. Businesses can ask themselves the following questions to understand how ready they are for AI agent adoption.
- How does an AI agent align with their overall business goals and strategic objectives?
- Do they have a governance framework in place to manage the deployment and ethical use of an AI agent?
- How does the AI agent help in enhancing existing processes and enterprise strategy?
- What strategies does the enterprise have in place to identify, assess, and mitigate risks associated with the AI agent?
- How will the enterprise manage the organisational changes required for the adoption of the AI agent?
- What key performance indicators (KPIs) will be used to measure the success and impact of AI agent initiatives?
- Is the current IT infrastructure capable of supporting the computational demands of the AI agent?
- Does the workforce possess the skills to use an AI agent, and what are the implications for talent acquisition and upskilling?
- How will the enterprise address security and compliance concerns related to the deployment of an AI agent?
Table 1 compares AI agents with traditional software.
| Characteristic | AI agent | Traditional software |
| Behaviour | Proactive, learning and adapting | Reactive and does not learn from user behaviour |
| Business logic | Machine learning- and LLM-based;
uses a reasoning-based approach to decide the steps needed to achieve a goal |
Predefined rules and workflows; follows hardcoded, rule-based logic without deeper reasoning |
| Frameworks | Operates on frameworks like LangGraph, AutoGen, CrewAI, TensorFlow and PyTorch that support dynamic learning and adaptation; these frameworks are flexible and modular | Operates on frameworks and architectures like JADE, Cougaar, FIPA-OS;
these frameworks are static and rigid |
| Communication | Dynamic function calling; leverages advanced APIs; utilises NLP to understand and generate human language, enabling smooth and intuitive interactions | Performs manual tasks; leverages standard APIs;
rule-based systems for communication, which can be rigid and less intuitive |
| Autonomy | Highly autonomous; capable of reasoning and decision making | Limited autonomy; follows pre-programmed rules |
| Decision making process | Can plan and break down complex tasks into smaller steps for better solutions | Direct logic with no real reasoning capabilities |
| Scalability | High | Limited |
| Learning | Continuous learning and improvement | No learning capability or limited learning within a fixed set of responses |
| Tool access | Can access APIs and databases to enhance responses | Does not access external tools or systems |
| Problem solving | Handles complex, multi-step problems by combining reasoning with external resources | Handles simple, well-defined problems with scripted responses |
| Response generation | Iterates on responses by gathering more data and refining the solution until accurate | Provides an immediate response without revisiting or improving the result |
| Complex query handling | Capable of solving highly complex or ambiguous queries by using multiple resources | Best suited for straightforward, well-defined queries |
| Use cases | Customer service, predictive analytics, automation, multimodal interactions | Data entry, basic reporting, routine operations, single-mode interaction |
Table 1: AI agents versus traditional software
The key benefits of AI agents are:
- Automate repetitive tasks to save time and manual work
- Analyse data and identify patterns quickly
- Operate 24/7 with no downtime
- Scale applications and adapt to increased demand
- Maintain consistent performance
Types of AI agents
There are several types of AI agents, and each is designed for specific tasks and environments.
Simple reflex agents
These agents make decisions based on the current perception. They follow condition-action rules and are suitable for observable environments. They have no learning skills or memory. Automated hand sanitiser dispensers are an example. These devices dispense the sanitiser when they detect motion, ensuring hygiene without human intervention.
Model-based reflex agents
These agents handle partially observable environments by maintaining an internal state based on perception history. Smart IV pumps are an example. These pumps adjust the flow rate of intravenous fluids based on real-time patient data, such as heart rate and blood pressure, ensuring safe and precise delivery.
Goal-based agents
These agents have goals and choose actions to achieve them. They consider the current state and the desired goal to make decisions. They are centred not only on existing but also on future conditions, and on the relationship between conditions and operations. Personalised treatment planning systems in healthcare are an example. These systems analyse patient data, including genetics and medical history, to create customised treatment plans for managing chronic diseases.
Utility-based agents
These choose actions that maximise the expected utility. An example is resource allocation systems in hospitals. These systems optimise the use of hospital resources, such as beds and medical equipment, to maximise patient care efficiency and minimise costs.
Learning agents
These agents learn from their experiences and improve their performance over time. They start with basic knowledge and adapt through learning. For example, in predictive analytics for patient monitoring, these agents learn from historical patient data to predict potential health issues, allowing for early intervention and better patient outcomes.
Hierarchical agents
These agents decompose tasks into subtasks and solve them hierarchically. They can handle complex tasks by breaking them down into simpler, manageable parts. Hospital workflow management systems are an example. These systems break down complex tasks into subtasks, such as patient admission, treatment scheduling, and discharge planning, to streamline hospital operations.
Key characteristics of AI agents
AI agent systems possess several key characteristics that distinguish them from traditional AI systems.
Autonomy
AI agent systems operate independently. They can make decisions and execute actions without human intervention.
Adaptability
By learning from experiences, AI agents can adapt to new environments and scenarios.
Goal-driven
They are designed to achieve specific objectives or goals.
Context awareness
Agent systems can understand and respond to the context in which they operate, allowing for more nuanced and effective interactions.
Proactivity
They can anticipate future needs and take proactive measures to address them, rather than simply reacting to external inputs.
Technical capabilities
These systems can leverage a variety of tools to enhance their functionality and handle complex scenarios.
Internet access: Allows agents to retrieve real-time information, perform web searches, and gather data from online sources.
Code interpreters: Enable agents to execute and interpret code, facilitating complex computations, data analysis, and automation tasks.
API calls: Allow agents to interact with external services and systems, enabling seamless integration with various platforms, databases, and applications.
Communication
Effective communication is a hallmark of an AI agent, enabling them to interact seamlessly with humans and other AI systems.
Scalability
AI agents are designed to scale efficiently, handling increased workloads and complexity as needed.
Robustness
AI agent systems are resilient and can maintain performance even in the face of challenges or disruptions.
Ethical considerations
These systems are built with ethical guidelines to ensure they operate responsibly and transparently.
Personalisation
AI agents possess the capability to retain individual preferences, enabling personalised interactions. They also have the capacity to store and utilise knowledge.
Integration capabilities
They integrate with existing systems and processes, enhancing overall organisational efficiency and effectiveness.
How to go about adopting AI agents in your business
Organisations must take a few critical steps to ensure successful deployment and integration of AI agents into their business.
Governance
Establish clear governance frameworks and compliance protocols to manage the deployment of AI agent systems. Define the policies for data privacy, security, and ethical use of AI.
Identify business use cases
Identify the business challenges that require attention. Also, understand the business benefits of AI agent adoption that are critical for the success of the enterprise. Select the targeted use cases and perform the proof of concepts (POCs) that can deliver the desired business and operational outcomes. Build value through improved productivity, growth, and new business models.
Change management
Implement change management to help teams to adapt new AI-driven workflows. Establish clear communication, stakeholder engagement, and support systems to ease the transition.
Ethical values
Address ethical considerations related to the use of AI agents, such as bias, fairness, and transparency. Implement ethical guidelines and review processes to ensure responsible AI deployment.
Infrastructure
Ensure that the existing IT infrastructure can support the deployment of AI agent systems. This involves upgrading hardware, software, and network capabilities to handle the increased computational demands.
Upskilling
Reskill employees to improve productivity by conducting various training courses and encourage them to perform POCs. Also, based on the role and skills of employees, identify the skill gaps and train them effectively to contribute to the enterprise transformation initiatives.
Risk management
Develop strategies to identify, assess, and mitigate risks associated with an AI agent. Implement robust monitoring and auditing mechanisms to track AI behaviour and ensure accountability.
AI agent reference architecture
Figure 2 shows the contextual architecture of an AI agent, giving its key components and layers.
Input data sources
Data sources provide the insight required to solve business problems. These are structured, semi-structured, and unstructured, and come from many sources such as user interactions, real-time data streams, and multi-model data covering images, text, video, audio, etc.
AI agents
AI agents process the input data coming from various data sources. The data is formatted in a way that it is interpretable by the LLM. The AI agent analyses the response and chooses between delivering the response directly to the user or using specific tools to perform additional actions, based on the LLM’s output. If the output involves data retrieval, the agent makes API calls, processes data and converts it into coherent responses.
Large language models (LLMs)
LLMs are a type of AI system trained on a large amount of text data that can understand natural language and generate human-like responses. The processed input is fed into the LLM, which generates a response based on its training.
Tool integration
This helps connect the agent with external applications, databases and automation tools to extend its functionality. Key aspects of tool integration include:
API integration: Agents communicate with other software systems.
Third party integration: Agents integrate with NLP systems and ML models to enhance its capabilities.
Automation tools integration: AI agents can automate repetitive tasks to increase efficiency.
Memory
AI agents remember previous interactions, user preferences, and ongoing tasks to provide a personalised and effective user experience. Key memory management characteristics include scalability, privacy, consistency and adaptability.
AI orchestration
AI orchestration involves managing the coordination and interaction between multiple AI agents to achieve specific goals or tasks. The key components of AI orchestration are:
Adaptive task management: System dynamically assigns and reassigns tasks to AI agents based on their capabilities, availability, and current workload.
Multi-agent collaboration: AI agents work together, sharing information and resources to achieve common goals.
Performance monitoring: System continuously monitors the performance of AI agents to ensure they are meeting the desired outcomes and standards.
Data repository
This includes comprehensive data covering both structured and unstructured data sources. It facilitates efficient data management by utilising both centralised and distributed repositories, employing vector stores for quick information retrieval, and leveraging knowledge graphs for contextual reasoning. The data is categorised and organised so that it can be used by an AI agent and models.
Output
This consists of AI insights that are transformed into personalised, context-aware results. These results are continuously updated. The system’s knowledge base is also refreshed in the process.
Governance
The AI architecture integrates essential governance and safeguards to ensure safety, compliance, and ethical AI deployment. It ensures AI agents operate safely, securely, and within regulatory boundaries.
Real world use case: Automated claims processing with AI agents
Use cases of AI agents are endless and evolving continuously, and businesses are experimenting with different ways to incorporate them. Here’s a scenario where an AI agent helps streamline claims processing operations (Figure 3).
Claim submission
A policyholder visits their healthcare provider for a medical service. After the service, the healthcare provider submits a claim electronically to the enterprise covering details like patient information, treatment provided, and costs incurred.
Initial claim review
An AI agent receives the electronic claim and begins the initial review.
Data verification
The agent cross-references the submitted claim data with the policyholder’s insurance plan details and medical history. It verifies the validity of the claim by checking for any inconsistencies or discrepancies, such as duplicate claims or services not covered by the policy.
Fraud detection
Using advanced machine learning algorithms, the AI agent analyses the claim for any signs of potential fraud.
Policy compliance check
The AI agent ensures that the claim complies with the policyholder’s insurance plan terms and conditions.
Approval and reimbursement
If the claim passes all the checks, the AI agent automatically approves it. It calculates the reimbursement amount based on the policy terms and initiates the payment process.
Escalation for manual review
If the AI agent identifies any issues or uncertainties during the review process, it escalates the claim to a human claims adjuster for further investigation.
Continuous learning
AI agents use feedback from the manual review process to continuously improve their algorithms and decision-making capabilities.
In summary, AI agents enhance the efficiency, accuracy, and overall effectiveness of the claims processing workflow in a healthcare insurance setting.
Open source platforms and frameworks for developing AI agents
The leading open source frameworks that provide the necessary tools and mechanisms to build, manage, and orchestrate AI agents are:
LangGraph
This framework is designed for building and managing multi-agent systems. It provides tools for defining agents, managing state, and orchestrating complex workflows.
CrewAI
This open source collaborative tool/framework is used for web searching, data analysis, and content generation. It supports dynamic task allocation and collaboration among multiple agents. Agents can review and improve each other’s output.
AutoGen
This is a framework used for building multi-agent systems with high customisation. It supports containerised code execution for complex tasks and simulations, and integrates with various LLMs.
OpenAI Swarm
A framework used for the orchestration of multiple AI agents, enabling them to work together on complex tasks.
LlamaIndex
This leading data framework is used for building LLM applications. It is used for data integration and retrieval. It provides data connectors that help LlamaIndex agents to seamlessly access and process external data sources, such as PDFs, Google Drive folders, web pages, SQL databases, and more.
Pydantic AI
This is a Python-based framework for building production-ready AI agents for data validation and serialisation. It supports OpenAI, Anthropic, Gemini, Ollama, Groq, Mistral, etc.
Dify
This no-code platform for building agents is user-friendly and easily accessible to non-technical users. It provides multi-model support covering GPT, Llama and Claude, etc. It connects with popular AI models and supports integration with external tools like Zapier, Make, etc, and provides strong data security.
Factors that need to be considered when choosing an open source platform and framework for developing AI agents are:
- Rapid prototype development
- Easy customisation
- Ease of use
- Better data integration
- Better NLP capabilities
- Cloud integration and scalability
- Capability to support large datasets
- Workflow automation
Many people perceive AI agents as merely enhanced chatbots, but they represent the future of the digital workforce. The use of AI agents across enterprises is becoming more and more widespread as they provide superior flexibility, intelligence, and efficiency over traditional software. They can autonomously execute tasks, adapt to new situations, and deliver more personalised interactions, making them an invaluable asset for modern businesses.
Disclaimer: The views expressed in this article are that of the author. Tricon Solutions LLC does not subscribe to the substance, veracity or truthfulness of this opinion.
