That is where Behe comes in because he proposes cases where it pretty much has to work like that such as blood clotting. And his question is why haven't these things been broken down into a possible sequence of mutations. You don't have to explain exactly how it happened, but rather how it could have happened gradually.
Functionally you can't assume that you get to conserve "right" answers without explaining the mechanism that keeps them. You are proposing something other than natural selection. It is pretty easy to point out that right answers in the end are very often wrong answers all along the way to that end. So you are left in a situation where natural selection has to throw away right answers to make evolution work, yet it has to keep them to make evolution work.
The situation is especially critical at the level of the cell. A line of feathers down my back, while suboptimal, might not keep me from reproducing. But badly functioning cells certainly will.
And according to other biochemists--presumably most of them, since Behe's criticism hasn't carried the day--those case don't have to work like that. Google Ken Miller, who's written a lot about how Behe's examples don't hold up. (And Behe's responded, and Miller's responded, etc. etc.) Miller's arguments seem pretty cogent to me, plus I've never been much for arguments from ignorance.
It is pretty easy to point out that right answers in the end are very often wrong answers all along the way to that end.
But they're very often right answers to something. Like, feathers could be the right answer to staying warm before they're the right answer to flying. And, of course, a lot of things might not be the answer to anything until they combine with something else.
That's one of the mistakes anti-evolutionists often make: thinking that an organism gets one chance at developing the right mutation, and that only one individual develops it. Consider, on the other hand, the possibility that say 30 percent of a population has a given mutation at any one time. It's not harmful, so it's not selected out; but it's not particularly helpful in the current environment, so it's not selected for. It's neither a right nor a wrong answer--it's just there. But if the environment changes, it could become part of a right answer. Or 20 percent of the population develops another mutation, and the 10 percent that have both do better for some reason. It's not always a binary right vs wrong situation.
First of all, Behe's claim that blood clotting suddenly sprang into being in its fully functional form is simply wrong. What you need to understand is that every component of the blood clotting cascade had another function before it changed to become part of the clotting system. In order to understand the evolution of blood clotting, you need to examine each component of the system. The structure and behavior of the von Willebrand Factor are not very unusual within a biological system: there are countless proteins that self-assemble upon various triggers such as a change in pH or a chemical signal. We also have countless proteins that change in conformation (shape) upon some sort of signal. The behavior of proteins is an unavoidable consequence of them being physical molecules that respond to physical stimuli.
It also is not unusual for cells such as platelets to catch each other and form little colonies; for them to do this upon sensing a specific signal is also not unusual. This ability evolved long before complex multicellular organisms appeared: the slime mold, Dictyostelium, is a fascinating organism that lives as single cells or forms a body, all dependent on external stimuli.
Okay, now that we've established that the two major players in the blood clotting system use fairly common mechanisms for their function, all we have to do is propose a likely sequence of mutations that led to them cooperatively working together to clot blood. At this point, there are many different possible sequences of mutations, and I can generate a multitude of testable hypotheses. If I were a graduate student interested in writing my dissertation on some aspect of the evolution of blood clotting, this is the point where I would hit up PubMed and start reading up on the work already done in the field. Then I would identify an area where very little work has been done (competition among scientists is brutal: I don't want to spend 4 years in research and then be unable to publish because I got scooped), propose an overall working hypothesis, and start generating a whole bunch of little working hypotheses that keep me busy in the lab from week to week.
Behe's idea of "irreducible complexity" is not meant to be scientific: it's meant to discredit scientists, and to discourage people from even examining science too closely.
Functionally you can't assume that you get to conserve "right" answers without explaining the mechanism that keeps them. You are proposing something other than natural selection. It is pretty easy to point out that right answers in the end are very often wrong answers all along the way to that end. So you are left in a situation where natural selection has to throw away right answers to make evolution work, yet it has to keep them to make evolution work.
The only mechanism that determines whether the "right" answers are "conserved" is reproduction. Selection doesn't play that much of a role (evolution proceeds whether selective pressures exist or not; it's a consequence of the highly mutable nature of DNA). Any organism that survives to reproduce is successful.