Adrian Stelmach is the founder and CEO of EXPLITIA.
getty
Thanks to the rapid development of new technologies in recent years, modern business is undergoing a clear transformation. Many industries are already implementing simple chatbots, and their capabilities continue to surprise us.
Today, however, more advanced solutions are coming to the forefront, including AI agents. These systems not only answer questions but also independently carry out assigned tasks. The move from passive tools to active, autonomous operational units is clearly redefining efficiency and workforce structures in modern enterprises.
One of my clients from the automotive industry was struggling with a high number of human errors in production planning and difficulties in predicting potential production slowdowns. Planning was based on a schedule provided by the end customer. In just-in-time production, however, it's important to remember that such a plan changes constantly, involves many variables and is continuously updated throughout the process. When scheduling is done manually, the time pressure is high, and even minor mistakes can lead to significant errors.
The standard system implemented at the beginning, based partly on linear regression models, didn't prove effective due to the large number of variables and the changing nature of the process. After analyzing, we concluded that the best solution would be to implement an AI agent to selectively automate selected parts of it.
The biggest implementation challenges involved properly synchronizing data sources so that the data would be available at the same time and could be linked to a specific production order. Another challenge was integrating the agent with the systems and procedures already operating in the manufacturing plant. The implementation required extensive change management, in-depth process analysis and adaptation to the factory’s existing standards.
Assigning some tasks to the agent, including error checking, identifying planning conflicts, building demand forecasts and making predictions, made it possible to significantly improve the quality of finished products, accelerate work and clearly reduce the number of errors.
Despite the enormous interest in this technology, market data is clear: We're still only at the beginning of the journey toward the full implementation and understanding of these types of solutions. According to a Camunda report, as many as 71% of corporations have already tested AI agents. However, only 11% of those companies decided to move to full-scale implementation.
This significant gap suggests that moving from the pilot phase to stable business use remains a major challenge.
The low percentage of full-scale implementations shows that organizations need to prepare much better for the adoption of AI agents. Effective implementation requires not only access to technology but also:
When defining the system architecture, it's necessary to determine where automation should take place and where the process should still be carried out by a human. Make sure to test how far the agent should go in delivering data and clearly define the areas covered by its tasks. In my experience, in many situations, external partners still want to communicate with a human. That's why, in many processes, a human should still play a standardizing role.
For example, a human collects data from external partners, defines its scope and then passes the data to the agent, which checks, verifies and schedules it, while also identifying potential errors and forecasting possible risks.
As with any AI system, an agent is only as good as the data it uses. To successfully implement an AI agent, the database itself must be properly built and the agent must be granted access only to the data you actually want it to use. Simulations should also be carried out to reliably verify whether the data provides full context.
In the case of one of my clients, there was too little data, which resulted in incorrect outcomes. Running human-in-the-loop simulations based on the scope of the person’s work and the data used in the process, and ultimately expanding the data scope, helped effectively eliminate the problem. That's why, at the initial stage, it's important to verify agents’ responses and analyze what may cause potential errors.
Long before implementing agents, you should communicate the change clearly, create space for questions and address any concerns that may arise to reduce employee resistance. It's worth preparing separate strategies tailored to each situation.
I know of a case in which a company needed to detect risks in order to respond to them in advance. The existing team didn't have time to perform this type of analysis. The introduction of agents made it possible to carry out without placing an additional burden on the employees.
Properly communicating the assumptions, goals, benefits and alternatives meant that, despite concerns related to the technology and its novelty, resistance didn't appear. On the contrary, employees were satisfied because they understood that without agents, these additional, time-consuming tasks would fall on them.
Practice also shows that even seemingly successful implementations can end in disaster if key elements are omitted. One of the best-known examples is the case in which Replit Agent deleted a production database during a code freeze, despite having been explicitly prohibited from making such changes. Although the database was eventually recovered, the incident received wide media coverage and likely did little to strengthen trust in solutions of this type.
AI agents are a solution that can significantly increase companies’ operational capabilities. By moving away from reactive chatbots toward proactive agents, we create space for better use of the human resources we already have. Although currently only around one in 10 companies fully uses their potential, it's precisely this group that may set the standards for tomorrow's digital economy.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
此内容由惯性聚合(RSS阅读器)自动聚合整理，仅供阅读参考。 原文来自 — 版权归原作者所有。