We present a symbiotic hybrid intelligence system for automated pick and place in a warehouse logistics setting. This domain is characterized by the need for systems to deal with long-tailed distributions of scenarios, consisting of a wide variety of relatively rare edge cases. Handling of these cases has great impact on the reliability, robustness, and commercial viability of these systems. Rather than attempt to learn a sufficiently generalized method to handle any potential unknown situation, our system focuses learning on the most common scenarios and leverages human flexibility to generate responses to the rarer ones. As more tail cases and their human-generated responses are seen, they are learned, moved into the body, and new tail cases take their place.