If a string has a length of 16 characters, but it is also known that the first 2 characters are fixed, and that the rest of the string has random letters but letters only, no digits nor special chars, would it still take years to crack a sha256 on these conditions or is it unsafe?
The fixed two characters known won't detract or compromise the security of the other 14 characters so let's just look at that that.
Looking at lower case alphanumeric only 26^14 is 6.4509975e+19 which is astronomical and in my opinion safe (this discussion around entropy, or better known as the study of randomness, has lead to the new opinion that long hard to guess passwords built up from phrases is better than 7 characters with numbers and special characters).
If the author want's assurance, you could consider adding a salt. To summarise I don't think these are unsafe conditions :)
The first two characters are fixed so it does not add or take away from entropy. Assuming that the next 14 letters are random, size of password space = 52^14 (lower and upper case letters). It is equivalent to hashing log(52^14)=14*log(52) = around 80 random bits.
I don't consider it secure against modern ASICs, when bitcoin network crosses 200 Exa hashes per second (about 61 bits). Even if just 10% of this power attacks you, it will only take about two months to crack it (my calculated estimate for maximum time. Average is about 1 month.)