Cognitive services and conversational digital assistants are emerging as the engine that powers natural interactions between humans, software services, devices and "things" - supported by advances in AI and human computations. Not surprisingly, many large and small tech companies are rushing to occupy this space by providing platforms for building cognitive services and conversational bots. Digital assistants interact in a natural way (through text or voice) with both software and humans to get information and perform actions, from checking the weather to booking a restaurants and a cab ride, managing cloud resources, answering simple scientific questions, and preparing a decaf latte using IoT enabled coffee machines. User requests or tasks are often expressed in natural language, an interaction ensues to clarify the intent and the details, and the answer is sought - or the appropriate service or device is invoked - based on the cognitive service understanding. While the potential of this new wave of services is exciting, it also brings significant challenges: we are far away from the comfort of developing deterministic software that responds to API calls by invoking other APIs. Now we have to understand, guess, explore options, take decisions based on probabilistic models over a large set of possible intents and services, all while engaging with users. Doing so brings a large set of engineering challenges related to the development, training, tuning and evolution of such services. This panel will discuss such challenges and identify interesting opportunities for research as well as promising trends.