Hi,
I haven't tried out ARCore yet, but it was reported that the precision is far less than Google Tango. But precision wouldn't be the main problem. The main problem with Google Tango etc. was the relocalization. Basically, those systems are 'visual odometry' systems - they accumulate the distance traveled. Because errors accumulate by visual odometry systems, these systems must be able to relocalize at certain places to reduce the accumulation error to zero again.
And the Google Tango relocalization (and probably the ARCore too) didn't work well with different lighting conditions. For example, once it was trained with objects (house, trees etc.) in the cloudy morning, it failed to recognize them in the evening or in sunlight. And without a robust relocalization errors accumulate and you get a precision like normal GPS (2-4 meters) with large garden environments.
I think the ultimate goal would be to robustly recognize already seen obstacles and use them to relocalize. But that's another story... (Deep learning cameras, cloud-based training, ...)
Regards,
Alexander