It's a play on the phrase "All models are wrong, but some are useful" in reference to statistical models (but often applied more generally to scientific or mental models).