Author: Mykhailenko A.
Technology is constantly transforming businesses, and today we are witnessing yet another such transformation driven by artificial intelligence. According to experts at McKinsey, the next phase of AI evolution is likely to significantly impact its usage and open up opportunities for tackling even more complex, multi-step tasks. This is the concept of AI agents, or “AI agents,” that will unlock the full potential of generative artificial intelligence (GenAI).
- AI Agents: What is This Technology and How Did it Emerge?
- Opportunities and Benefits of Agent-based AI
- How AI Agents Work
- AI Agent Capabilities for Business: Key Scenarios
- How to Work with AI Agents
AI Agents: What is This Technology and How Did it Emerge?
Artificial intelligence is already everywhere: in apps, smart cars, and online stores predicting our actions. Most of the wow-moments from interacting with AI have come from GenAI, which became accessible thanks to ChatGPT. However, agent-based AI is not exactly a new technology — the concept of agents was described by Alan Turing, and chess-playing programs created in the 1960s demonstrated the workings of typical agents making decisions in a limited environment.
What is agent-based AI? At IBM, it refers to programs that autonomously perform tasks on behalf of a user, plan steps to achieve a goal, and use available tools. Larry Hämäläinen, a senior partner at McKinsey, uses the term to describe “software entities that organize complex workflows, coordinate actions by applying logic, and assess responses.” Agents can either fully automate specific processes (Larry believes their contribution to this will reach 50% by 2055) or complement human workers, as GenAI already does, by diving deeper into tasks.
The renewed interest in AI agents surged in 2023 when tech giants like Google and Microsoft began investing in the technology. Chinese IT giant Baidu is also working on AI agents, while Meta (Facebook) already offers its LLAMA framework for creating AI agents. Sam Altman, founder of OpenAI, believes, “By 2025, we may see the first AI agents become real workforce tools, significantly changing company outcomes.”
Opportunities and Benefits of Agent-based AI
AI agents are synonymous with intelligent automation, which is already partially happening thanks to generative AI. However, beyond automating complex tasks that typically require human intervention, AI agents open up many other opportunities due to their unique capabilities:
- They understand natural language as instructions.
- They provide complete, accurate, and personalized responses tailored to specific users.
- They build on previous results at each subsequent step of solving a task.
- They deliver high performance (especially multi-agent systems).
- They work with a wide variety of tools and digital ecosystems.
- They can gather data through physical interfaces (sensors in IoT ecosystems).
- They quickly adapt to changes in task conditions.
These abilities allow AI agents to be used across various industries and deliver measurable results:
How AI Agents Work
Like GenAI, Agent-based AI is also built on large language models (LLMs). However, while traditional LLMs are typically limited to a specific set of knowledge and conclusions, agents autonomously use the available tools to search for and analyze the necessary information. Furthermore, they aim to adapt to user requirements and preferences, retain all past interactions, and can plan future actions without human intervention.
The operation of an AI agent can be schematically represented as follows:
Although AI agents are autonomous in decision-making, the goals, tasks, and environment for their operation are still defined by humans. One team designs and trains the system, another deploys it and provides an intuitive interface for interacting with the agent, and finally, the end user sets the specific goal and determines the tools to be used to achieve it.
If an AI agent lacks knowledge, it will refer to other agents, databases, or use web search. For example, if a user asks the AI agent to choose the best time to fly to Paris for a vacation, the agent will study airline schedules and find acceptable options based on price and arrival time. To minimize the risk of changes in plans, the agent will autonomously set the next task: to examine archives and weather forecasts to determine when the weather conditions will be most favorable. By consolidating data from different sources, the AI agent will offer the user several options for flight and arrival dates.
If we compare AI agents to traditional generative models, such as OpenAI’s ChatGPT, the picture would look as follows:
| AI Agent | Generative AI (ChatGPT) | |
| Positioning | An assistant who doesn’t just follow commands but also makes decisions to help achieve the set goal. | A tool for creating something: for example, text, images, or videos. It generates human-like text well based on the input data or instructions. |
| Training | Uses continuous learning algorithms and adaptive models, taking previous experience into account when solving subsequent tasks. | It learns from the provided database but encounters difficulties in handling new requests that go beyond the training data. |
| Autonomy | Performs tasks independently. | Requires user prompts to generate responses. |
| Task Management | Breaks down complex tasks into subtasks and solves them sequentially. | Generates content based on the input data, handling only one task at a time. |
| Integration | Can interact with various applications and APIs. | By default, it does not integrate with other applications. |
| Example of Use (Bank Customer Support Service) | Suitable for complex tasks: for a client seeking a loan, it will review their credit history, suggest filling out an application, send it to a specialist, monitor the review process, and return with the final decision. | Suitable for generating answers to frequently asked questions such as “What are the bank branch hours?” or “What is the balance in my account?” |
AI Agent Capabilities for Business: Key Scenarios
AI agents have not yet become a commonplace tool in business, but this could change in the near future. Check out these four examples to evaluate the potential uses of this technology.
Credit Scoring in Financial Institutions
One agent can handle communication with a potential borrower to gather as much information as possible for decision-making. Another will collect the necessary documents and check them for errors. A third will analyze the movement of funds in the borrower’s accounts. And a fourth will consolidate this information and propose a final decision on the application.
Resource Optimization in Healthcare
Similarly, several AI agents can segment patients in the emergency room, creating queues based on the reason for the visit. They can also create preliminary treatment plans or manage medication administration. This will help address the shortage of healthcare workers and improve patient service quality.
Improving Marketing Campaign Effectiveness
A marketer can provide an AI agent with information about target customers, describe the marketing campaign goals, and outline the intended communication channels. The agents will then develop and test different strategies using customer analytics, communication history, and market research. Some agents will assist in creating personalized content (both text and visual), ensuring alignment with the brand book, and evaluating risks to the company when implementing unconventional ideas.
Managing Technical Debt
Multi-agent systems can enhance the development and optimization of applications. Some agents will check the code for compliance with requirements, others will perform QA tasks, a third group will focus on threat analysis, and a fourth will ask questions to developers to accelerate goal achievement and improve task accuracy. The modernization of legacy applications and the reduction of technical debt remain significant challenges in IT: according to McKinsey, more than 70% of software used by Fortune 500 companies was developed 20 years ago.
How to Work with AI Agents
There are three ways to work with AI agents: through super platforms, software shells, and individual user tools. Super platforms are the next generation of business applications where AI agents are one of many features available as needed. Software shells work with third-party services through APIs, while custom agents are developed specifically for a company by fine-tuning a pre-trained large language model (LLM). The first option is easier to scale, while the last is more flexible, allowing for custom configurations. For example, agents can be created for a call center, where they answer customer questions using the company’s customer data, past conversation records, and internal guidelines.
AI agents are still in the early stages of development, and their full potential has yet to be realized. However, businesses should start preparing for the deployment of AI agents, understanding that achieving results will require additional resources, both computational and human. For instance, the new system will need to be trained and extensively tested before it can operate autonomously. In addition, businesses will face other challenges related to AI agents, such as data privacy protection, identifying false or biased responses, and technical difficulties.
You can rely on the Colobridge team to implement the most complex IT solutions. Contact our experts to learn how to effectively utilize available AI solutions and prepare for the next phase of their development—AI agents.
