You're thinking of the inferior system that predated T-9. The whole point of T-9 was using a language-specific word frequency list so that words can be entered with just one digit per letter.

The main word pairs that this often fails for are "me/of" and "no/on" (edit: other super-common words are "go/in", "he/if", and "up/us"; "am/an" isn't a problem since you usually write "I'm"); prefixes that can end with "-er" or "-es" are also ambiguous. For those you need to press the arrows to select a possibility manually. If there's an ambiguity for longer words (usually, if the word you're looking for is not in the dictionary), you might have to use the arrows ahead of time and then keep typing before the space; it will try to complete suffixes that go after that prefix.

There are only around 2000 ambiguous digit sequences after excluding the -er/-es pair, and for most of those at most one is likely to be used in texting even if they're "common" in English.

The most ambiguous sets, with 6-8 "common" "words", are:

  2253 able bake bald bale cake calf
  22537 abler bake[rs] balds bales cakes calfs
  2273 acre bard bare base cape card care case
  22737 acres bards bare[rs] base[rs] cape[rs] cards cares cases
  24337 aides bides cheep cheer chefs cider
  269 any bow box boy cow cox coy
  4663 gone good goof home hone hood hoof
  46637 goner goods goofs homes hones hoods hoofs inner
  7243 page paid rage raid sage said
  727733 parred parsed passed rapped rasped sapped
  7277464 parring parsing passing rapping rasping sapping
  729 paw pay raw ray saw say
  7327 pear peas reap rear sear seas
  74337 ride[rs] rifer sheds sheep sheer sides
  7627 roar robs snap soap soar sobs
  7673 pope pore pose rope rose sore
  (no +7 since "popes" isn't common)
  78337 puffs queer ruder ruffs steep steer
  7867 pump puns rump rums runs stop sums suns
  787433 purged pushed rushed stride strife surged

Think about how often, while texting, you actually use the second-most-common of a given word set.