Deep learning is still not production ready.

Maybe 90% of the results are great, but the other 10% are not only 'not great' but plain wrong and sometimes even dangerous.

And I think gp posts is also showing that the last bump to make it production ready is still not taken and might even be impossible to take.