Our natural languages are just too confusing to talk about this topic.
I think GP meant that when we say that a decision happens "inside the head" it means that the brain performs a computation where it combines all the "training* the brain got during its life so far and the perceptions of the current events and produces stimuli for muscles that will act on the world.
In order to produce the right kind of stimuli for the muscles the brain performs a modeling of what the future will look like after the muscles have been moved.
Imagine a scenario where you're walking with a child on the side of a road with some car traffic. You hold the child's hand.
The brain constantly asses the sensorial input of pressure in your hand and gauge what will happen if the child hand slips from yours. The brain computes the possible outcomes of that particular child veering off in the traffic and weighs in the horror of an accident that might happen into the decision making of whether to move the muscle of your hand just a little bit more so that the child's hand doesn't slip away.
This computation, the imagining of the accident happening and your reaction may or may not reach your awareness (i.e. feel conscious). It involves an enormous amount of assumptions about how the world works and about the things you value in life, your past experience, your expectations of torment in the case something bad happened while the child was under your watch.
This "computation" is determined by millions of factors that we colloquially smuggle under a single deceivingly simple word: "you"
All those parameters, what it means to be "you", is constantly changing based on what happens every single second to you. Everything that makes you you is based on this constant flux of input your brain processes, since the day you were born. The structure of your body as built by your genes also affects this since it produces the very machinery that processes these inputs and reacts to them causing new inputs to be fed, since the day you were born.
This incredibly complex mechanism, that involves constant imagining and predicting the future, weighting different predicted outcomes and reaping the rewards of correct ones and suffering the consequences of bad ones happens continuously in your head without any "agency" on your part. This very mechanism is your agency.
Whether each little step in this cascade of events that implements your "agency" is "deterministic" it doesn't change the fact that the computation that defines that agency must happen for the agency to have effects on the real world.
A deterministic evolution of a system still requires the system to exist. When you predict the exact movement of a billiard ball hitting other balls and the walls of the table you don't say "ah since I know exactly where the ball will go this means that the other balls and the table are useless, I'll just place the ball I care about in the final position and call it a day, since it's all predetermined anyway*. You don't say that because the positions of the other balls after the action will matter for the next actions.
Similarly, even if the underlying laws of physics are deterministic and can be in principle used to predict the behaviour of a complex system like brain, the act of predicting it must perform at the very least the same computation that the brain is doing (and much more, since that one brain constantly reacts to other things including many other brains making their own action/reaction loops?)
Even if we all lived in a simulation it would make sense to say that your brain makes the decisions. Because that's a useful way of talking about what this organ does.
It's a useful way to coarse grain reality in systems and environment. It's compatible with another way of looking at the overall simulated system from the POV of the simulator. Those two points of view can coexist and are equally real.
In practice our real world is so complex that even if in principle the laws of physics are deterministic, you need a pretty special being, (a laplacian deamon) with enormous compute power to carry out the predictions of the whole system. This being would have to be external to the system it is predicting and better have to effect on it otherwise it would have to simulate itself too.
Whether a laplacian daemon exists, or whether the underlying laws are indeed deterministic, doesn't change the fact that it's useful to describe individual agents in the system and talk about them in terms of what these agents so based on their local inputs and outputs.
Both views are "compatible" with each other. In the same way as the view of a gas being made of molecules that move and bump to each other each with their own position and momentum is compatible with the way of talking that the gas has a temperature and a pressure.