We investigate how explicit symbols, structured representations, and linguistic ideas can help capable models coordinate and continue their work. One starting point is the communication passed from one model call to another.