Experiment 1 demonstrates that problem solving knowledge can be applied while a move is in progress in certain Tower of London (ToL) problems. A two-stage move process is often delayed in the second stage when participants have been misled by similarity to a previous problem. We suggest this is indicative of misgivings about the chosen move caused by on-going analysis of the move that is being made. Experiment 2 swapped the stages of the two-stage process and again reported more hesitancy in the second stage when participants had been misled. We conclude that it is desirable for models of problem solving to evolve so that they can apply the same learned problem solving knowledge both before a move is selected and while the move is being made. We then describe a model of ToL problem solving that fulfills these criteria and has been computationally-implemented within an embodied cognitive architecture.