Logging in software development plays a crucial role in bug-fixing, maintaining the code and operating the application. Logs are hints created by human software developers that aim to help human developers and operators in identifying root causes for application bugs or other misbehaviour types. They also serve as a bridge between the Devs and the Ops, allowing the exchange of information. The rise of the DevOps paradigm with the CI/CD pipelines led to a significantly higher number of deployments per month and consequently increased the logging requirements. In response, AI-enabled methods for IT operation (AIOps) are introduced to automate the testing and run-time fault tolerance to a certain extent. However, using logs tailored for human understanding to learn (automatic) AI methods poses an ill-defined problem: AI algorithms need no hints but structured, precise and indicative data. Until now, AIOps researchers adapt the AI algorithms to the properties of the existing human-centred data (e.g., log sentiment), which are not always trivial to model. By pointing out the discrepancy, we envision that there exists an alternative approach: the logging can be adapted such that the produced logs are better tailored towards the strengths of the AI-enabled methods. In response, in this vision paper, we introduce auto-logging, which devises the idea of how to automatically insert log instructions into the code that can better suit AI-enabled methods as end-log consumers.