What problem it solves
Earlier soft forks, such as BIP-34, used a rising version number in block headers to track miner support. BIP-9 explains why that did not scale: it supported only one change at a time, required coordination between proposals, and did not allow permanent rejection, since as long as one soft fork was not fully rolled out, no future one could be scheduled.
How BIP-9 works
BIP-9 reinterprets the block version field as a bit vector in which each bit tracks an independent change. Each soft fork deployment has a name, a bit chosen from a defined range, a start time and a timeout. Miners who support the change set its bit in the blocks they produce. Those signals are counted for each retarget period.
The stages
A deployment moves through states. It is first defined, then starts after its start time, then enters a signalling period. If enough blocks in a period signal support, the deployment locks in, waits for another period and then becomes active, after which the new rules are enforced. If the median time past reaches the timeout before it locks in, the deployment is considered failed. The BIP also describes a fallow period after which the bit can be reused.
Why it matters
Version bits make it possible to roll out several upgrades in parallel and to measure support in the open. They also give nodes and users a clear signal of how close an upgrade is. Whether miner signalling should be the only deciding factor has been debated, and later activation methods have been used for some upgrades.
What it means for you
Normally nothing is needed from you except keeping your software updated when an upgrade is announced by the project through its official channels.