Traditional automation is based on the premise that programs will operate according to the program, making it easy to control quality, making it ideal for use in corporate operations. However, it is difficult to respond flexibly, and humans are always involved in decision-making, so no system has yet reached the level of "leaving decisions to humans."
Because generative AI cannot know the output until it has actually generated it, it is difficult to control the quality, which leads to repeated retries and the risk of lowering labor productivity. Prompt engineering may enable generative AI to generate output close to the expected level, but because it ultimately depends on human skill, it is difficult for an organization to guarantee the quality of its work.
AI agents have the autonomy that traditional systematization and generative AI lack, so they can be tasked with larger groups of tasks than ever before. This significantly reduces the number of points of human intervention, including important decision-making that could not be entrusted to traditional systems, and is expected to improve labor productivity for anyone who uses them.
One issue facing AI models in general, not just AI agents, is the monetization of the companies that provide them. Currently , subscription-based pricing is the norm, but the costs of complex calculations and the massive amount of power consumed rise the more users use the model, making it highly likely that this will eventually fail. In order to provide sustainable services into the future, a new billing system is required.
Report format: PDF (8.2MB)
Original data: PowerPoint, 73 slides, A4 size