The success of integrating artificial intelligence into the corporate environment directly depends on companies' ability to extract and formalize the tacit knowledge of their employees. This refers to those practical skills, intuitive decisions, and expert judgments that accumulate over years but are rarely documented in files or instructions. This concept, rooted in Michael Polanyi's philosophy ("we know more than we can tell"), is now becoming central to the corporate AI race.
Why universal models stall on specific tasks
Universal language models show impressive results in areas with open data and clear criteria. However, when it comes to a company's internal context, specific business processes, and subjective expert assessments, written instructions alone are insufficient. This resembles Frederick Taylor's attempts in the early 20th century to systematize the experience of factory workers—a labor-intensive process that is not always effective.
Bridgewater case: expertise as fuel for AI
A striking example is the recent study by Bridgewater AIA Labs and Thinking Machines Lab. The teams tested leading models (Claude Opus, Gemini, GPT) on six real-world financial tasks. With simple instructions, accuracy ranged from 45.6% to 50.1%. However, when Bridgewater experts prepared specialized prompts reflecting their own judgment, the metric soared to 74.3–78.2%. Automatic optimization of instructions yielded no additional gains—the human factor proved decisive.
The culmination was fine-tuning the Qwen3-235B model on data labeled by Bridgewater analysts. Result: an average accuracy of 84.7%, a 29.8% reduction in errors compared to the best universal model. Task processing costs dropped by a factor of 13.8. This approach, termed "differentiated intelligence," involves calibrating AI to the unique processes of a specific organization.
Ford: salvaging departing expertise
Meanwhile, Ford demonstrates another aspect of the problem. Over three years, the automaker hired, promoted, or brought back about 350 experienced technical specialists. Their task is not just to work but to transfer their knowledge: review designs, identify hidden defects, train young engineers, and participate in configuring automated quality control systems. Ford Vice President Charles Poon stated directly: "Artificial intelligence is only as good as the information used to train it." The company acknowledged that it had previously failed to systematically preserve the knowledge of departing veterans, and now it must make up for lost ground.
My analysis
The market is beginning to grasp a fundamental truth: AI will not replace experts, but it will become their ideal assistant if we learn to "extract" knowledge from them. The key asset of the future corporation is not so much algorithms as the ability to systematize and digitize the unique experience of its key employees. Companies that solve this problem will gain a tremendous competitive advantage. Those that continue to rely solely on "raw" data and generic models risk falling behind.