Longer feedback loops are inherently much harder to steer.

The way we learn is by taking an action with a predicted effect, seeing how the outcome matched the prediction, and then adjusting.

The longer it takes from action to effect, the longer the feedback loop cycle, and the slower the learning cycle.

The more action/effect cycles you get, the more you absorb the knowhow of how to steer this particular thing.

But the longer the feedback loop, also the more diffuse and indirect the effects of the steering, and the harder to extract the learning.

If there's a short feedback loop with a very direct connection, then the outcome arises straightforwardly out of almost entirely your action.

But if the loop is longer, there's more environmental factors and outside factors that affect the system from your action to the outcome.

Those muddle the analysis; was the effect because I steered it wrong, or because of that random gust of wind halfway through?

In longer feedback loops you'll tend to oversteer and then have to steer back to get it back on track.

In the limit this can cause oscillations around the intended level.

It's harder to extract the signal from the loop.

Long feedback loops are orders of magnitude harder to learn how to steer.