This is a good piece.
I think election models can be useful for trends and to add color to the uncertainty inherent to polling. And add more objectivity than pundits.
But yeah the certainty some people have with them and how consumers interact with them are probably net harmful.
I like how FiveThirtyEight presented their forecasting performance across politics/sports/etc on their site in this regard. It helped show how it is really a small sample environment in politics given the limited number of elections.
I really recommend reading the model methodologies. You'll realize how there are a lot of assumptions that go into these models.
For that reason, I tend to construct an ensemble forecast of all the models.