Here is an "all-matricial" explanation:
$$K:=\pmatrix{-K_1&K_{21}\\K_{12}&-K_2}, \ \ P:=\pmatrix{a_1&a_2\\b_1&b_2}, \ \ D:=\pmatrix{\lambda_1 & 0 \\ 0 & \lambda_2}.$$
The set of 4 equations you have given can be expressed into a single matrix equation, up to a sign change:
$$\pmatrix{a_1&a_2\\b_1&b_2}\pmatrix{\lambda_1 & 0 \\ 0 & \lambda_2}=\pmatrix{-K_1&K_{21}\\K_{12}&-K_2}\pmatrix{a_1&a_2\\b_1&b_2}.$$
In other words:
$$\tag{1}PD=KP \ \ \iff \ \ D=P^{-1}KP,$$
in which we recognize the classical diagonalization relationship that is currently used in the solution of linear differential systems with constant coefficients.
Recall: Let us write the given differential system under the form $X'=KX$. Using change of variable $X=PY$, it becomes:
$$PY'=KPY \iff Y'= \underbrace{P^{-1}KP}_{D} Y$$
If we have chosen $P$ such that $P^{-1}KP=D$, (see equ. $(1)$) where $D$ is a diagonal matrix, we get two "decorellated" very simple differential equations $x'(t)=\lambda_k x(t) \ \ \Rightarrow \ \ x(t)=C_k e^{\lambda_k t}.$