The pads are randomly generated, not random. Each pad needs to be longer than your message. No part of the pad is ever reused; if you have the first half of a pad, you can decrypt anything encrypted with that half, but it tells you absolutely nothing about the second half of the pad, because it's all random, not an algorithm. If you have the cleartext, you could not reverse-engineer the pad from it, and even if you could, you couldn't use that to determine the rest of the pad.
Here's how it works.
You generate a random pad. In the old days, when the term originated, it was literally a pad of paper with random letters.
The sender and receiver must have identical copies of each pad. For example, lets say you generate a pad for each day of a year, and distribute a copy to each embassy. So each pad has a master number, 1 through 365, and each embassy has it's own in that series.
Each and every one of those copies must be physically secure. If they are, the communications are unbreakable. If they are not, the communications are not.
In the above example, each day's pad might be on, in this day and age, a secure USB key, shink wrapped, with anti-tamper foil. If, in daily inspection, any key is missing, appears to have been altered in any way, the shrink wrap scratched or warped, whatever, every embassy is immediately directed to burn their copy of that pad. Once the pad is used, or at the end of the day, each copy is burned to prevent accidental reuse.
It's part of a cryptographic system, not a complete methodology in and of itself. And it's no different than the idea that, say, public/private key encryption is secure until you misplace your private key.