There is definitive proof that it doesn't work. But because a lot of people use it there is a tendency for bollingerv bands, and support levels etc. to be helpful
It depends on the person doing the TA though. After watching Jesse's hangout call,with DaveT and following his posts I believe that Jesse uses TA to get excellent results.
My belief is that TA works when it works and then doesn't work when it doesn't.
Basically, enough people practice TA that in "quiet" times, TA can dominate the trading. It can help explain group behavior when the group is being predictable in its psychology. But around news events that alter the perception of the business, it is of much less use. It can help determine how far something runs or falls. It seems to me to be best when there's very little else to go on.