I certainly don't think polling is useless, but polls do miss.
Roughly ~80% of races will be called correctly from polling alone. That's much better than a coin flip, but it means ~20% of races will be called incorrectly. And excluding primaries, average polling error across national and state level races is about ~5.4 points. And in general historically, polling tends to slightly overestimate Dems, though not always.
2020 had slightly worse polling error than average, but significantly overestimated Dems compared to average. 2018 conversely had below average polling error and slightly overestimated Republicans. 2016 had about average polling error, but overestimated Dems. When you look at 538's Deluxe forecast the model expects polls to overestimate Dems this cycle, so keep that in mind when looking at the model.
There is also variation in polling error by state. In recent cycles, polling has tended to miss the most in Ohio and Wisconsin, but polling in Georgia and Arizona has been spot on conversely. Polling error by state isn't always consistent either. Florida had very accurate polling in 2016, but a relatively large miss in 2020.
Don't be surprised if a candidate that was polling well loses, that can happen from time to time with this level of polling error, especially if a race hasn't had much polling.