Can someone explain this for me please?

I am doing a factorial anova analysis,
and I have this factor A with two levels: off/on.

Now in the response, I know A should explain a lot.

However I can also measure some other properties, covariates, which I know are related to A. So these covariates are high when A is OFF and low when A is ON.

If I include these in the model as covariates,
A barely explains shit.

If I dont have them, A explains at least half of the data.

So how does this work?

Does this mean the covariates are just as good / better at explaining the difference in the response as A?