Oh, that is what you are getting at. We had this many many times before: If the system you need to realize can meaningfully be modelled by several mostly independent logical strands of execution, an RTOS will help you sort out the control flow. If all your system needs to do can be modelled via a linear deterministic sequence of actions (eg read sensor->communicate sensor to host->sleep->repeat), you do not necessarily need an RTOS (although parts of that sequence, eg the host communication, may bear some hidden concurrency, in which case an RTOS, again, is the tool of choice).
Most current embedded systems generally serve more than one logical strand of execution, so there are very few use cases of single-threaded embedded systems.
The other issue you mentioned was hard real time. Admittedly, very few systems require true hard real time, but one of the benefits of RTOS over more complex operating systems is that they allow you to design your system such that an upper bound can be guaranteed for most time critical operations (that is the traditional definition of real time).
Please consult with one of the many many resources that discuss RTOS before putting up such elementary questions on this forum. You can also browse previous threads here and will find that the basics and rudimentaries of RTOS have been explained here several times before.
Thanks.