A cognitive assistant that allows a maintainer to speak to an application using natural language is disclosed. The maintainer can quickly interact with an application hands-free without the need to use complex user interfaces or memorized voice commands. The assistant provides instructions to the maintainer using augmented reality audio and visual cues. The assistant will walk the maintainer through maintenance tasks and verify proper execution using IoT sensors. If after completing a step, the IoT sensors are not as expected, the maintainer is notified on how to resolve the situation.