TL;DR in bold:
We don't have crystal balls to predict where technology will take us, but the purpose of cryptography is to develop algorithms that have just this kind of resistance on a very fundamental level.
Mathematically speaking, in terms of "honestly" brute-forcing a single plaintext from a single ciphertext, knowing everything that would otherwise be needed to decrypt the plaintext other than the key, a 256-bit block cipher should be considered uncrackable by the laws of thermodynamics; our Sun will not emit enough photons in its active life (before it becomes a black dwarf) to enable a computer that works at the electron level with perfect efficiency to try every possible key in the 256-bit keyspace. The computer would be starved for energy after trying about 2^218 key values, meaning that if it tried keys perfectly randomly without replacement, and the key was also chosen perfectly randomly, it would only have a one in 275 billion chance to have found the correct key, given roughly 5 billion years before the Sun goes nova.
There is, however, one distinct disadvantage to the use of any computational hardware or software in existence today; it is already obsolete. It would be foolish to assume that any binary implementation currently used to encipher messages, nor any computer, language or even logical structure used to develop or execute that binary code, will still be in existence in 100 years. We already have problems with the availability of hardware that can read or write storage devices that were universal standards only 20 years ago. Even power plugs have changed fundamentally in the last 50 years, and those plugs and receptacles (and the exact specs of the power they use) are only regional standards as it is. Hell, 100 years ago we were riding horses, not driving cars. So, if you plan to encrypt something and then store the device used to produce it, I can virtually guarantee that in 100 years everything about the device will be so obsolete it will be unusable.
Given this 100-year expected timespan, I would rely on one of the simplest and yet most secure ciphers known to man; the one-time pad.
Simply create a numeric "alphabet" relating a number value to every character symbol you wish to be able to encipher (don't forget any needed punctuation, such as spaces, commas, periods, single/double quotes etc; you must not use more than 100 symbols, but most fully-formatted English-language messages will be well underneath this limit, even with capital letters). Then, gather a series of numbers, from zero to 99. You will need one of these numbers per character of the message. These numbers must be genuinely random; serious users of one-time pads usually gather bit data from environmental sources, such as measurements of background radiation detected on a radio telescope. For today's purposes, you could purchase a 100-sided die and give it a roll around the inside of a bowl to produce each number. It should also be acceptable to use a CSPRNG, provided that it is properly seeded from a source of true entropy (provided that the seed value is discarded after use, you probably won't need enough numbers from the PRNG for an attacker to be able to predict them). Make sure the memory is fully cleared from any electronic device used to generate random data.
To encrypt, take the random pad, the numeric alphabet, and the message you wish to encrypt into a place where you can't be seen by anyone else or any surveillance equipment (you know, just in case). Take the first character of the message, look up its alphabet code number, then get the first number from the random pad, and subtract the random number from the character code. If you end up with a number less than zero, subtract that number from 100. Look the resulting number up on the alphabet chart, and write down the character as the first character of the ciphertext. Repeat with the second character of the plaintext and the second number of the random pad. Continue through the message until you have encoded every character of the plaintext message. You must have used every number of the random pad, in order, once and only once.
When you're done, you burn or otherwise completely destroy your copy of the plaintext, and physically secure the random pad and alphabet. This is important; nobody should be able to get their hands on the random pad for 100 years, but they must be able to get to it in 100 years. The alphabet is not technically a secret, but it's inclusion with the ciphertext is a clue as to the use of a one-time pad (leading people to go searching for the pad), so I'd recommend keeping it with the pad. The ciphertext itself is perfectly secure in its encrypted form; you can chisel it into the stonework of the Supreme Court building if you wanted (and could do it fast enough to finish before the DC Metro Police took you away). It could be any of a number of alphabet ciphers, lending a little entropy to mask the discovery that it's a OTP in the first place.
To decrypt, someone must have the ciphertext, the same one-time pad and the same alphabet table that you used to encrypt. They take the ciphertext's first character, look up the number in the alphabet table, then add the number from the one-time-pad and modulo by 100 (simply truncating the hundreds place wherever it shows up). They turn the resulting number back into a character using the alphabet, and write it down. They then continue, character by character and number by number. Again, they must use every number of the random pad, in order, once and only once.
The one-time pad is a very old system, first being described in 1882 and re-invented in 1917, and it has the ultimate advantage; it is provably impossible to crack. Being genuinely random, there is no mathematical pattern to any of the numbers of the random pad (which forms the "key" of the cipher), and because each number is only ever used once (if you have more than 100 characters you will see the same two-digit number in the pad twice, but you can never predict when), the pad itself is never repeated. Thus, any attempt to decipher the message without the exact same pad is futile; it could produce gibberish, or it could produce a message of exactly the same number of characters, but whose content is something completely different. The only way to be sure that every character of the message is decrypted correctly is to have the exact same sequence of random numbers that encrypted the thing in the first place.
The disadvantages that keep it being used more universally are that it ideally requires a pad of infinite length (to handle any number of messages of any length), and there's a large possibility of offset error; someone can miss some or all of a message sent by the other person, so one person has crossed off fewer numbers on their pad, resulting in all further messages between the two parties being indecipherable and no real way to recover without exchanging a new pad. In your case, these are moot, but pretty much every other cipher system invented in the last 100 years has attempted to mitigate these disadvantages (primarily the infinite-length key) while trading as little as possible of the ideal security of the system.