The Wikipedia page on $E_7$ tells me:

Even though the roots span a 7-dimensional space, it is more symmetric and convenient to represent them as vectors lying in a 7-dimensional subspace of an 8-dimensional vector space. The roots are all the 8×7 permutations of (1,−1,0,0,0,0,0,0) and all the $\begin{pmatrix}8\\4\end{pmatrix}$ permutations of (½,½,½,½,−½,−½,−½,−½) Note that the 7-dimensional subspace is the subspace where the sum of all the eight coordinates is zero. There are 126 roots.

This presentation is indeed more symmetrical than anything 7-dimensional I could have come up with myself. But more importantly it makes computations really easy because, magically, the Killing form is just the restriction of the ordinary inner product on the ambient 8-dimensional space to the 7-dimensional subspace where the roots live. (Wikipedia does not state this explicitly but it can be easily checked from the list of simple roots corresponding to the nodes of the Dynkin diagram that Wikipedia is kind enough to list.)

The question in the title ('where does this come from?') is really two questions:

**1) A reference request:** if I use this presentation of the root system to simplify my computations, who do I give credit to?

**2) A more 'philosophical' question:** where does the nice presentation in terms of a bigger space 'come from', mathematically?

The idea of realizing a degree $n$ root system in the $n$-dimensional 'sum of coordinates equals zero'-subspace of an $(n+1)$-dimensional space is of course very familiar: it is how we normally describe the root systems of type $A_n$.

But in the $A_n$-case the appearance of the extra dimension seems very natural. Thinking about the root system as coming from the Lie algebra $\mathfrak{g} = \mathfrak{sl}_{n+1}$, we can either argue that the $(n+1)$-dimensional space 'surrounding' the $n$-dimensional Cartan subalgebra is `really' the Cartan of the central extension $\mathfrak{gl}_{n+1}$ of $\mathfrak{g}$ or simply accept that everything there is to understand about $\mathfrak{g}$ can be seen inside its 'defining' representation, which happens to be $(n+1)$-dimensional. (I say it a bit sloppy but you hopefully get what I mean.)

It seems that neither of these explanations is available in the $\mathfrak{g} = \mathfrak{e}_7$ case. It definitely does not have a non-trivial 8-dimensional representation (defining or otherwise) and I also never heard about any interesting central extensions.

So is there another explanation of this '$A_n$-like' behavior of $E_7$? The best I could come up with is that the nice 8-dimensional presentation of the $E_7$-root system is 'inherited' from the $A_7$-root system sitting inside of it. But as far as explanations go this feels like cheating since I only found out that there is an $A_7$ sitting inside $E_7$ by looking at the very description of the root system I am trying to explain!

extendedDynkin diagram. Consulting, e.g., Bourbaki shows that the extended Dynkin diagram of $E_8$ has an extra node on the 'long' end. Deleting the penultimate node ($\alpha_8$, in Bourbaki's numbering) gives the $E_7 + A_1$ diagram. You may be interested in a similar recent question by @MatthiasKlupsch. $\endgroup$2more comments