We all know that there are various encoding schemes to encode data like Manchester Encoding, before sending it. But why is there a need to encode the data to a signal? Why can't we just transmit the waveforms?
Electrically speaking, encoding schemes are important to keep the line balanced - even if the data changes state often enough to clock and be distinguishable from a dead line, if it has an uneven number of 1's/0's the receiver develops an unbalanced offset voltage. Encoding ensures both that there are no long strings of 1's or 0's, and that the total number is effectively balanced over time.
The systems simply work better with the benefits brought by various encoding schemes.
So your Manchester Encoding example has the benefit of ensuring frequent -- more frequent than would be the case with typical data -- voltage changes on the wires. More frequent voltage changes have the side effect, in this example, of making clock/timing work better. Other encodings have other benefits that befit their transmission mediums.
Traditional encoding mechanisms were built to deal with certain media characteristics - specifically electrical issues on traditional copper media or high BER carriers in the various wireless domains. With modern transmission systems there has been a push to streamline these mechanisms to simplify protocol stacks and make better use of available bandwidth by removing associated overhead (i.e. removing SONET encoding from the equation for IPoWDM).
In the larger sense the issue with transmitting native waveforms with no encoding is that some kind of system is required to map between the information to be transmitted and the media over which transmission is to occur. In the case of a T1 circuit (for example) this corresponds to a set of voltage thresholds and an expected signalling rate. Without such a system there is no guarantee (or even likely hope) that what one side transmits is what the other side receives.