Whenever you send a packet in TCP, the recipient sends back an "ACK" packet confirming receipt. If some timeout interval passes, the packet will be sent again. This means that some packet loss will not interfere with a TCP connection.

This is used for congestion control: Imagine you have a computer with a 1gbps LAN hooked up to a 100mbps network. You can send packets at 1gpbs over the LAN, but 90% of them would be dropped at the speed change point. The TCP stack will notice it is losing packets, and will throttle itself accordingly.

TCP is actually more complicated than just this, but this is one of the more important parts.

Packets can be dropped and TCP will retry (among other things; packet loss is a hueristic.) It won't just cut the session on some amount of packet loss, that would lead to constant failures.