10/100 autosensing is for backwards compatibility with old hosts that are 10M only.
10M was the only ethernet for a long time. When 100M was introduced, it was extremely expensive and only used for backbones or servers (and not all servers either, just the important one or two that could justify the cost). Therefore it was assumed that 10M devices would be around for a long, long time.
Interestingly enough: 10/100M NICs were so expensive that using one to talk to a 10M device was initially considered wasteful. Why spend all that money to talk to a 10M device unless that 10M device was scheduled to be upgraded. If it was decided NOT to upgrade
Why was 100M either so expensive initially? Besides the fact that new tech is usually more expensive at first, at the time the only other network technology that was 100M/s was FDDI which was extremely expensive and extremely difficult to manage. So, the 100M/s ethernet vendors could charge high prices and it would still be cheaper than FDDI. If bread cost $100/loaf and you found a way to make it for $10/loaf, you'd charge $99/loaf for as long as you could, right?
I remember a Cisco salesperson in 1996-ish saying "we expect 10/100 switch ports to drop to $100/port eventually, so in the future we won't offer 10M... just 10/100M." At the time this seemed like crazytalk.