Using the most powerful AI models for all tasks is no longer the most efficient option for companies, especially when most daily operations are repetitive and do not require exceptional abilities. This is what was stated in a recent report by the research company Gartner, which explains that about 90% of the steps performed by artificial intelligence agents fall into normal tasks, such as information classification, engagement tools and data inspection, which are tasks that open-weight models can handle with efficiency close to closed models. Economically, the report shows that the cost of these models can reach a quarter of the cost of the cheapest closed model that provides similar performance, and the difference can be much greater compared to using advanced models at every stage of the work. However, the actual savings are still lower due to the possibility of retries and increased output length. But the most important aspect is not only the cost, but the control and protection of the data. Companies' reliance on external model interfaces means sending parts of their data and operational context to external parties, making open models a strategic choice for those seeking to retain more control over their technical architecture. Gartner predicts that companies will rely on open models to perform 90% of process agent steps by 2030, a sign of a potential shift in how enterprise AI systems are built. This reflects a new trend in the competition, where the preference will not necessarily be for the most powerful model, but for a system capable of intelligently dividing tasks and choosing the most suitable model for each process, with the continued role of advanced models in research and development and more complex tasks.