They sit inside the testing,
red-teaming, and data work that many AI systems still need, so more AI deployment should send more work their way. But they do not own the model or the customer outright, so big clients can automate, insource, or switch faster than they could with a true platform owner.