The paper addresses the problem of modelling an itiformation flow in manufacturing agents (subsystems) consisting of people, machines, robots, etc. It matches very frequent real-life situations in which the following should be answered. How to structure an exchange of information between the elements of an agent? What is better, complete itiformation but lleavily delayed, or incomplete information less delayed? The formal mathematical model can be used to generate some examples of knowledge that is useful in answering the above type of questions. Quantitative models of an information flow evaluation, however, are often too complex to serve as the tools useful in the knowledge retrieval process. This realization, also observed in other areas of applied mathematical modelling, has led to increasing interest and research in non-quantitative modelling. In the paper, the application of signed directed graphs in the process of knowledge acquisition for autonomous agents is investigated and illustrated with the case of itiformational balance.