2025 is called the year of AI agents. Are AI agents really a must-have for everyone, and what AI assistants are the Ukrainian divisions of EPAM, Intetics, Levi9, P2H and Railsware working on?

Pavlo Yalovol, VP of Innovation at Intetics

Progress in the field of generative AI is so rapid that it makes sense to focus not on years, but on quarters, and sometimes even months.

Yes, the topic of Agentic AI gained popularity back in 2024, and already in January 2025 OpenAI makes it possible to create agents without leaving ChatGPT. At Intetics, we started creating AI agents for our clients back in the summer of 2024 and are currently continuing to work on new agents that:

• measure the dimensions of car windows;
• create autotests based on a text description;
• search for POIs (Points of Interest) in a given region and periodically update data on those that are outdated,
• create proposals for our clients and automate the execution of internal operational tasks.

We believe that by 2025, the popularity of agents will significantly decrease as we move closer to AGI (Artificial General Intelligence), and people will no longer need dozens of agents created to perform a single task tied to a specific industry. Rather, each user (or organization) will have their own AI assistant that can help with any task, taking into account the context of a specific person or organization.

Vadym Vlasenko, Solutions Architect at EPAM

2025 promises to be a pivotal year for AI agents, which are rapidly gaining popularity due to their ability to automate a variety of tasks.

Over the past year, we at EPAM have been actively developing AI agents for areas such as e-commerce, software development, contact centers, marketing, insurance, the tourism industry, etc.

I work with my own AI agents every day to help with research and documentation. For example, they can independently collect data from the internet and integrate the results into systems like Confluence, which saves a lot of time.

Additionally, our team has created a proprietary no-code platform for building AI agents that allows for easy integration with various integrations, such as databases and project documentation. This platform opens up automation to a wide range of users, allowing them to quickly adapt agents for business tasks without deep technical knowledge.

This confirms the growing trend of integrating AI into business processes, driving increased efficiency and the ability to focus on strategically important tasks.

Igor Kozlov, Data & ML Engineer at Levi9

Undoubtedly, the last year can be considered a breakthrough year in the field of Generative AI. The speed and scale of development of both models and architectures of AI agents are impressive. One of the main areas of application of agents is customer support. Most requests can be processed by agents with a knowledge base, which significantly reduces the need for human intervention and optimizes the cost of maintaining customer support.

Last year, we implemented such a system as a POC for one of our clients. And this year, we are working on creating AI Buddy for internal use. This tool will help our employees in self-development and answer frequently asked questions about the company’s internal processes. Buddy is also implemented using agentic architecture, which allows us to improve the quality of answers, because each data source is responsible for its own agent and they can work in parallel.

In addition to customer support and self-development, it is also important to understand the impact of agents directly on the software development process. They can now not only help write code, but also do code reviews, optimizations, and even build applications from scratch (a good example of this is v0 by Vercel, https://v0.dev/).

However, it is important to consider security risks: how safe is it to transfer code and data to third-party models? For some customers, the optimal solution is to use open-source models deployed locally or in the company’s cloud.

At Levi9, we already use agent coding on certain client projects. This is governed by additional agreements that take into account all necessary security aspects.

Nika Tamayo Flores, Product Lead, Railsware

It may sound a bit provocative, but AI agents are not a panacea and certainly not a must-have for all products. Yes, their popularity is growing rapidly, and we also plan to work with them in the future. However, they are definitely not a primary need for some complex products, such as Coupler.io, which helps collect data and create dashboards.

Why? The main «trick» of AI agents is interaction through queries. But do many users really want to understand how everything works «behind the scenes»? Most likely, they are more interested in three things: query speed, price and quality of the result. And whether it works thanks to AI or other technologies is not a fundamental question, while AI is more of a nice bonus.

Classic machine learning is often enough to cover most of the needs of product users. It is cheaper, easier to implement, and, moreover, more environmentally friendly. If you haven’t thought about how much water and electricity is needed to execute a single query in an AI model, you can read here .

However, the impact of AI on human activities is already tangible and continues to grow. Artificial intelligence becomes especially effective in cooperation with humans and their intelligence. It can quickly analyze large volumes of data, while humans can make strategic decisions, taking into account the broader context. Such cooperation helps to solve tasks more effectively: employees are freed from routine, and the company receives a significant increase in productivity.

Dmytro Mashchenko, Chief Operating Officer, P2H

The demand for AI agents is growing in areas that require automation of complex but repetitive processes, such as web development, e-commerce, marketing, and customer support. In web development, AI enables the creation of more intuitive and interesting web products. It allows you to create content, analyze data from different angles, and make decisions based on it. But most importantly, AI optimizes the operational work of the team, speeding up the development cycle. For example, AI automates repetitive tasks such as code generation, testing, prototyping, and debugging. Thanks to this, developers can focus on more complex and creative aspects of web development. In addition to automating processes, AI can help improve the results of the work performed or suggest improvements based on the data it processes. AI can effectively automate aspects such as code quality, User Experience, and SEO.

More than a decade ago, we at P2N developed a method that allowed us to quickly calculate clear volumes and deadlines. For example, the team identified a set of standard elements for cutting, styling, and aligning according to the design in 12-15 minutes. At the height of the AI ​​era, we came to the conclusion that this method could also be taught to a machine. Thus, the AI ​​Tool was born, which simplifies the assessment of the complexity and cost of website development — of course, with developer moderation and result adjustments as needed. This tool combines our experience in creating websites with modern artificial intelligence (AI), allowing clients to quickly receive understandable results and plan projects more effectively. The tool analyzes the design or structure of the site and offers three assessment options: Classic front-end development (HTML, CSS, JS). Site creation with Elementor (WordPress plugin).

Website design on WordPress. The user uploads a design to Figma, screenshots of the site, or provides a link that is scanned by AI. The model counts unique elements, determines the complexity of the implementation, and provides three options with a budget and timeline calculation. We used the Supervised Learning method to train the model. It is trained to recognize design elements based on labeled examples. The model was improved until it began to produce results with 85% accuracy in just 2 minutes. The optimized architecture allows you to work quickly even with large requests. 2025 can be safely called the era of AI agents.

We see how automation of routine processes and implementation of AI become an integral part of business. Our tool confirms this trend: it actually performs the role of an «AI agent for evaluating web projects», automating a complex process and freeing up time for more important tasks. We do not stop there and strive to further increase the accuracy of the model, as well as expand its functionality and make it available for analyzing new design formats. At the same time, we are actively exploring the capabilities of AI agents for development in order to offer customers even more solutions that automate their business.