The successful implementation of artificial intelligence in the corporate environment directly depends on the active participation of employees in the training of specialized models. Financial Times columnist Sarah O'Connor draws attention to the critical importance of so-called implicit knowledge. These are practical skills and unique ways of making decisions that are shaped by years of experience, but are rarely recorded in official instructions. Universal neural networks show outstanding results only in those areas where there are huge amounts of open data and clear verification criteria. However, in tasks that require a deep understanding of the company's internal context and expert judgment, written regulations alone are categorically insufficient to achieve high quality.

A clear confirmation of this theory was a joint study by Bridgewater AIA Labs and the Thinking Machines Lab, published in late June. The teams tested the capabilities of artificial intelligence on six complex tasks from the investment company's real-world workflows. When using standard prompta, large universal models showed an accuracy of responses of only 45-50 percent. However, when Bridgewater experts prepared specialized instructions, this figure rose steadily to 78 percent. Moreover, the further training of the Qwen3-235B model on the data marked up by the company's specialists allowed us to achieve a phenomenal average accuracy of 84.7 percent. At the same time, the cost of processing similar tasks has decreased by almost fourteen times, which confirms the effectiveness of the differentiated intelligence approach.

Large industrial companies are also aware of this pattern and are changing their HR strategies. Over the past three years, Ford has purposefully hired or returned about 350 experienced technicians. Their main task is to check complex structures, identify potential malfunctions and directly train young engineers to work with automated quality control tools. Charles Poon, the company's vice president of hardware development, emphasized that the result of the algorithms depends solely on the quality of the information used for their training. The management admitted that it had previously not systematically preserved the invaluable experience of senior engineers, which now requires urgent adjustments to internal business processes for successful digitalization.