As the title says:
Do you individually animate every step or is it a loop?
Please have a look at the Blender documentation on Animation, IPO-curves (F-curves), and Actions … as well as Keyframing.
Now, here’s a 30-second snapshot of the idea so you won’t think I’m telling you to “RTFM.” 
Basically, an animator will set one or more “key poses” of their character, called “key frames.” In the good ol’ days, if you had a job at (Ink and Paint …) Disney, your job was probably to be an “in-betweener,” who had to draw all the “other” frames. Today, the computer does that, but the name “keyframe” stuck.
Each time you set a “key frame” in Blender (or any other animation package), the computer will snapshot things like the location, rotation, and scale (size) of the objects that you’ve identified. (Or the “bones” in an “armature.”) At these “key” frames, the position of the object will be exactly what you specified. But what about “in-between?” Well, the computer uses a type of mathematical curve which will smoothly connect each key-frame point, providing a good-looking intermediate value “in-between” each of them.
You certainly can specify that a piece of animation … a walk-cycle, for example … should loop. (Mathematically speaking, it’s easy: the curve can simply “wrap around,” and/or the set of keyframe points can be automagically repeated. Computers are great at math. 
) As an animator, there are several ways to do that. One way is by defining “actions” (such as “one step”), then telling the computer to repeat that action (and maybe, to blend it with other actions!) multiple times. The computer sorts it all out into these points and curves and even makes it look easy.
You can even do other things: give the computer “a physics problem” to solve, such that it figures out what to do. Or, you can give it “goals” and “constraints” to figure out. (Computers are great at doing dumb mathematical chores that we humans don’t like to do, and by-the-way doing them very fast.)
And … Blender has all :eek: of that. (Well, almost all of that. CG Nerds are always inventing new things.)
As a certain other group of computer nerds are fond of saying: "TMTOWTDI™ = There’s More Than One Way To Do It!" Blender supports many different ways to approach the same task.
Welcome to Blender, and to the addicting world of 3D! 
(If you think that “(yawn…) computers are just about web-sites …” … heh.™)
It’s always good if your just starting out to practice by building walk cycles…
if your goal is to build for games…
the walk cycles will be used quite a bit…
if your building for Animation…
you will find that only a small portion of the total work will need a walk strait up walk cycle…
as many scenes will need their own custom animated walk motion just for that scene…
because the character didn’t just take a step… they did a step and then a leap…
or a step and then a stoop… or a step and turned a sharp corner… or a step and a skip… or a step and twisted ankle fall…
or something… other than just a strait up walk cycle… (oh here’s this guy and he’s walking down the street type of thing)
Anywayz… hope that helps…
… and that is where the oft-overlooked power of “NLA” (Non-Linear Animation/Action Editor) can come into play.
This very powerful feature lets you, first, “individually describe,” and then, “blend together,” multiple ‘actions’ such as “basic walk-cycle, step, skip, and leap” in a very natural way.
“Digital computers being what they are,” many very-useful techniques are built upon the ideas of “digitally merging common behaviors, and then (maybe …) dealing with the exceptions.” After all, if you can manage to describe anything “in terms of if…then…else, one way or the other,” then the computer can help you solve it. Yes, given that “every scene, more or less, is more-or-less one of a kind,” the computer (Blender) still offers “an impressive set of labor-saving devices” . . .
I will be the first to admit I have only barly sratched the surface of using the NLA editor… but then most of the Animators I speak with don’t seem to make use of it either as most are trained in the old style of pose to pose Keyframe style animation… But I agree 100% with sundialsvc4 that the NLA seems like a much much better way to do it…