The most widely held, but rarely tested, hypothesis for the origin of animals is that they evolved from a unicellular ancestor with an apical cilium surrounded by a microvillar collar that structurally resembled present-day sponge choanocytes and choanoflagellates1,2,3,4. Here we test this traditional view of the origin of the animal kingdom by comparing the transcriptomes, fates and behaviours of the three primary sponge cell types, choanocytes, pluripotent mesenchymal archeocytes and epithelial pinacocytes, with choanoflagellates and other unicellular holozoans. Unexpectedly, we find the transcriptome of sponge choanocytes is the least similar to the transcriptomes of choanoflagellates and is significantly enriched in genes unique to either animals or to sponges alone. In contrast, pluripotent archeocytes upregulate genes controlling cell proliferation and gene expression, as in other metazoan stem cells and in the proliferating stages of two closely-related unicellular holozoans, including a colonial choanoflagellate. In the context of the body plan of the sponge, Amphimedon queenslandica, we show that choanocytes appear late in development and are the result of a transdifferentiation event. They exist in a metastable state and readily transdifferentiate into archeocytes, which can differentiate into a range of other cell types. These sponge cell type conversions are similar to the temporal cell state changes that occur in many unicellular holozoans5. Together, these analyses offer no support for the homology of sponge choanocytes and choanoflagellates, nor for the view that the first multicellular animals were simple balls of cells with limited capacity to differentiate. Instead, our results are consistent with the first animal cell being able to transition between multiple states in a manner similar to modern transdifferentiating and stem cells.