the cellular approximation theorem
When you build spaces out of cells, you want maps between them to respect that cell structure — to send the skeleton-by-skeleton scaffolding of one into the other. Most maps do not, on the nose. The cellular approximation theorem rescues the situation: every continuous map between CW complexes can be wiggled, without changing it up to homotopy, into one that does respect the cell structure. It is the technical workhorse that makes inductive, dimension-by-dimension arguments in homotopy theory legitimate.
Recall a map f: X -> Y of CW complexes is cellular if it carries the n-skeleton X^n into the n-skeleton Y^n for every n. The theorem says: every continuous map f: X -> Y between CW complexes is homotopic to a cellular map; moreover, if f is already cellular on a subcomplex A, the homotopy can be taken to fix A (a relative version). The mechanism is a dimension count: an n-cell, being n-dimensional, cannot fill up a cell of dimension greater than n, so by a smoothing/transversality-style argument any map of a low-dimensional cell can be pushed off the interiors of higher cells and compressed into the lower skeleton — formalized via the fact that a map S^k -> S^m with k < m is null-homotopic.
The payoff is that cellular approximation underwrites the foundations: it is a key lemma in proving the Whitehead theorem, in setting up cellular homology, and in the basic computation that pi_k(S^n) = 0 for k < n (any map of a small sphere into a big one is homotopic to a cellular, hence non-surjective, hence null-homotopic map). It also justifies treating maps of CW complexes combinatorially, cell by cell. An honest caveat about scope: 'cellular' is a homotopy-level improvement, not a strict one — the theorem gives a homotopic cellular map, not that your original map was cellular, and the homotopy genuinely moves the map; also the theorem is firmly a CW statement and says nothing about general topological spaces, where no skeletal structure exists to approximate into.
Why is pi_1(S^2) trivial? Give S^2 its CW structure (one 0-cell, one 2-cell) and S^1 the structure with one 0-cell and one 1-cell. Any loop S^1 -> S^2 is homotopic, by cellular approximation, to a map into the 1-skeleton of S^2 — which is just the single 0-cell, a point — so the loop is null-homotopic.
Cellular approximation squashes any loop on S^2 into the 0-skeleton, proving simple connectivity.
The theorem produces a homotopic cellular map, not a claim that your original map was cellular; the approximating homotopy really does move points. And it is strictly a CW-category result — there is no analogue for arbitrary spaces, which lack skeleta to approximate into.