The system x+y=10, 2x+y=16 is really just its numbers. Strip away the letters and keep a grid: [12111016]. Each row is one equation; the columns hold the x coefficients, the y coefficients and the constants. In matrix form it reads Ax=b with A=[1211] and b=[1016].
Row operations are the elimination moves you already know, done on whole rows. Subtract twice row 1 from row 2: [101−110−4], which says −y=−4, so y=4, and row 1 then gives x=6. Swapping rows, scaling a row by a nonzero number and adding a multiple of one row to another never change the solutions, because each turns true equations into true equations.
When A is square with detA=0, you can multiply by the inverse instead: x=A−1b. Here detA=1⋅1−1⋅2=−1 and A−1=[−121−1], so A−1b=[−10+1620−16]=[64], the same answer. Cramer's rule finds each unknown as a ratio of two determinants.
If detA=0, the methods that divide by it fail, and that failure is information. Row-reducing x+y=3, 2x+2y=7 produces the row 0=1, so the system has no solution: the two lines are parallel. A row of zeros, 0=0, would instead mean infinitely many solutions.
Think of it like this. An augmented matrix is a spreadsheet of the system with the letters deleted, and row operations are elimination moves applied to whole rows at once. Like tidying a spreadsheet, you work toward a clean layout, ones on the diagonal and zeros elsewhere, where the answers can be read straight off the last column.
Core idea. Write the system as a matrix equation and solve by row reduction, x=A−1b, or Cramer's rule.