This paper presents a new way to develop artificial intelligence (AI) controllers using Adams and Bonsai, a reinforced learning tool from Microsoft. Neural network AI are becoming more popular for controlling systems influenced by unquantifiable disturbances or that can’t easily be controlled by traditional methods. Reinforced learning differs from other types of AI training (supervised or unsupervised) in that it does not require a large amount of up-front data collection, from test or from simulation. Instead, the training is conducted interactively between the brain Bonsai Neural Network brain and the Adams system model. For each simulation step, the brain sends an action to the Adams model which is then simulated one step and returns the simulation states. From the change of the states, the brain draws a conclusion whether the actions were good or bad and adds this to the training history. As this process is very time consuming, all computations take place in Azure, the Microsoft cloud computing environment. Multiple simulations can be run in individual containers to speed up the training. A streamlined workflow between Adams and Bonsai has been developed that any user of Adams/Controls will be familiar with. It is very similar to other types of export of an Adams model and does export a functional mockup unit (FMU) as the core of the interaction. The user is then guided through the steps of creating a container image and uploading this to Azure. The training curriculum is described in an “inkling” file and a skeleton file with selected signal variables and training parameters is also created for the user, making the whole setup process very simple. After the training is successfully conducted, the trained neural network brain can be downloaded and be co-simulated with Adams in a local container or be installed on a hardware platform such as the Raspberry PI. This presentation will show example models with the different steps required to connect Adams and Bonsai, setup a training and finally download and deploy the trained brain with the original model to test robustness and performance. There will also be a discussion about future development of the Adams – Bonsai interface.