Glass sponge reefs on the continental shelf of western Canada and south-east Alaska are considered stable deep-sea habitats that do not change significantly over time. Research cruises using a remotely operated vehicle equipped with accurate GPS positioning have allowed us to observe the same sponges at two reefs in the Strait of Georgia, British Columbia to document recruitment, growth and response to damage over time. Spermatocysts and putative embryos found in winter suggest annual, asynchronous reproduction. Juvenile sponges (2–10 cm in osculum diameter) in densities up to 1 m−2 were more concentrated near live sponges and sponge skeletons than away (Spearman rank correlations, P < 0.0001 for live cover and for skeletons), suggesting that recruitment occurs in particular regions using sponge skeletons as substrate. Most sponges showed no change in shape or size over 2–3 years, but some had died while others showed growth of 1–9 cm year−1. Deposition rates of reef-cementing sediments were 97 mm year−1 at Galiano Reef and 137 mm year−1 at Fraser Reef, but sediments eroded so that there was no net gain or loss over time. Sponges recovered within 1 year from small-scale damage that mimicked bites by fish or nudibranchs; however sponges did not recover from crushing of a large area (1.5 × 2 m2) even 3 years later. These observations and experiments show that while recruitment and growth of sponge reefs is more dynamic than previously thought, the reefs are not resilient in the face of larger-scale disturbances such as might be inflicted by trawling.