Fundamentals
Every clock drifts a little every day. NTP is the quiet, mostly-invisible protocol that keeps millions of computers close enough in agreement that nobody notices.
Every computer has a clock, and every clock drifts — a few seconds a day is typical for an uncorrected quartz oscillator, which sounds trivial until you remember that TLS certificates, distributed databases, and financial transactions all depend on machines agreeing on the time within a much tighter tolerance than that. NTP, the Network Time Protocol, is the reason they usually do agree.
An NTP client asks a time server what time it is, and a naive implementation would just take the answer at face value — but the request and response both take real, non-zero time to travel across the network, and that travel time is exactly what would throw the answer off. NTP's actual trick is measuring the round-trip delay of the exchange and using half of it to estimate one-way latency, then adjusting the reported time by that estimate. It's an approximation, not a perfect correction, but it's a good enough one that NTP-synced clocks typically agree to within tens of milliseconds over the public internet.
NTP organizes time sources into numbered "strata." Stratum 0 devices are the actual physical time references — atomic clocks and GPS receivers, which get their signal from GPS time (see GPS Time vs. UTC vs. Unix Time for how that differs from UTC). Stratum 1 servers are directly connected to those reference devices; stratum 2 servers sync from stratum 1, and so on. Your laptop is typically syncing from a stratum 2 or 3 server, several steps removed from the actual reference clock, which is part of why NTP's error-correction math matters as much as it does.
A clock that's drifted enough will start rejecting valid TLS certificates (which are time-bounded), corrupt the ordering of events in distributed systems that rely on timestamps to sequence writes, and throw off any cron-based scheduling. NTP running quietly in the background — it's on by default on essentially every modern OS — is why this is rare in practice. When it does go wrong, it's almost always because NTP itself is blocked by a firewall or misconfigured, not because the protocol failed.
Related
One of NTP's own ultimate reference sources, and how it differs from the clock on your wall.
Another way a "correct" clock can still cause a production incident.
See the practical result of all this synchronization: the same instant, everywhere.