These days, any user can create a simple program, like a calculator: you just need to ask an AI to generate code for your task. AI assistants can also be used to create more complex products, such as a template landing page or website, but IT expertise is needed to verify this code and detect bugs timely. For truly complex, brand-new software or language models, you’ll need a big team of professionals with unique expertise in various IT fields.
Researchers from ITMO’s Institute for Artificial Intelligence have developed ProAGI, a multiagent system that doesn’t just generate code, but conducts the full cycle of industrial software development. The system can automatically design, develop, test, and improve software products by working as a team of several IT specialists: an architect, an analyst, a developer, a tester, and a technical consultant. The system can meet the needs of companies that need to develop proprietary software for professional tasks beyond the capacities of popular AI agents and don’t have the resources for more IT specialists.
“We have developed ProAGI, a multiagent system that can implement the entire software production cycle. You give the system an idea as a prompt and as a result, you get a ready-to-use product. This approach is what makes us different from competitors on the Russian market. They provide developers with access to proprietary language models and co-pilot tools that are integrated into the development environment and can assist with finishing specific parts of code. On the contrary, our system can accelerate the full development cycle and bring down its cost by up to 10 times, which makes modern IT solutions accessible to any sector of the economy. In the current year, we expect the system to significantly reduce the implementation time for projects at small (under 500 staff) IT companies – from 3-6 months to 1-3 weeks,,” shares Denis Nasonov, the head of the AI in Industry Laboratory and, a senior researcher at the Research Center “Strong AI in Industry.”
From the outside, ProAGI functions quite simply: users need only to describe their task in a prompt – then, the system will plan the work, collect data, design the architecture, write code, test its quality, and improve as needed. However, on the inside, ProAGI is a complex system. First, agents (autonomous intelligent systems that can interact with their environment and make decisions without a human’s input) clarify the task and its context, and analyze the requirements, limitations, and success criteria. The task is then divided into subtasks and subfields that are assigned to each agent; dependencies between agents are also decided at this step.
Next, agents design the solution’s architecture. For generating software code, agents draw knowledge and context from a company’s databases, libraries, and services. Based on these, they design the architecture and write tests and documentation for solution validation. The result is a solution ready for deployment and maintenance. During the development process, the validating agents built into ProAGI test the software for compliance to the required metrics and standards, analyze it, and correct the detected mistakes.
ProAGI can be used to quickly create reliable digital solutions in different fields. In telecommunications, it can design and support internal platforms; in industry, it can design digital twins, monitoring systems, or ERP modules to control a business’ resources. ProAGI can also be of use for medical companies and educational institutions. Medical companies can use it to create clinical registries for collecting and storing data on patients with specific conditions, while educational institutions can automate their LMS, a digital platform for creating and organizing courses and collecting learning analytics. Apart from that, ProAGI can be used to control expenses, manage assets, and conduct audits at financial companies, while governmental regulators can create information systems and analyze data.
In these fields, ProAGI has already been tested on pet projects and real-world initiatives. For instance, for Tatneft, one of ITMO’s partners, the system created software for generating dashboards and evaluating the parameters of pressure in deposits. Such analytics can help understand how wells perform in oil production. Additionally, ProAGI became the foundation of an HR application where agents automatically analyze applicant CVs, select fitting candidates, and conduct primary interviews in text form. Instead of manually searching for hours, HR specialists now get a list of candidates who are ready to communicate.
However, ProAGI isn’t limited by these fields – it can be configured to fulfill other business tasks, as well. The multiagent system supports corporate processes and integration with companies: it can connect to repositories, APIs, and databases to work locally without any external data transfer. Moreover, ProAGI uses its own knowledge (documentation, codes, and guidelines), further increasing its accuracy and corporate reliability.
At this stage, the developers are choosing a model for interaction with customers: B2B, B2C, or both. The team will also develop a preproject agent that could evaluate the difficulty of the developed solutions. Based on this evaluation, it would be easier to decide if ProAGI can solve the task fully on its own or develop a major part of software architecture that will then be completed by an IT team.
