Releasing a new feature traditionally meant shipping new code to everyone at once, with no easy way back if it misbehaved. Feature flags are switches built into the software that turn a feature on or off without another release, so a change can be revealed gradually or hidden instantly.
This allows showing a feature to a small group first, measuring the effect, and rolling it back the moment something looks wrong. The cost is complexity, since flags accumulate and must be cleaned up, but the control they give over risk is usually worth the discipline.