after many researches on the subject, I can't find any convincing argument anywhere, so I come to you about this problem which has been brought by some of my high school students.
Let $ABC$ be a random triangle and $I$ his incenter. It is known that $AB=c$, $AC=b$ and $BC=a$. I'm looking for a clean way to express the distance AI only from $a$ $b$ and $c$ parameters (no angles). When I searched on the internet, I found the formula : $$ AI^{2}=\frac{p-a}{p}bc$$ Where $p=\frac{a+b+c}{2}$ is the semi-perimeter. The formula is quite nice, but I can't find a proof. I tried with law of cosines, heron's formula, but I can't quite catch the idea which will bring me this particular formula. Any idea ?

I'm using $s$ for semiperimeter (that's more common).
First, drop perpendiculars from $I$ onto $a,b,c$ and call them $ID,IE,IF$ respectively. All are equal to the inradius $r$.
Then, it is well known that $AE = AF = s-a$. I provide a short proof below:
It is also well known that if $\Delta$ is the area of $\Delta ABC$ (by a slight abuse of notation, but it is commonly used), then $\Delta = rs$. Here's a proof:
Then, by Pythagoras' theorem in $\Delta AIE$ (since $\angle AEI = 90^\circ$): $$\begin{align}AI^2 &= r^2+(s-a)^2 \\ &= \Delta^2/s^2 +(s-a)^2 \\ &= (s-a)\left(\frac{(s-b)(s-c)}{s} + (s-a)\right) \tag{Heron's Formula} \\ &= (s-a)\left(\frac{2s^2 - sb - sc -sa +bc}{s}\right) \\ &= \frac{(s-a)bc}{s} \tag{as $s(a+b+c) = 2s^2$} \end{align}$$ QED