No level skips reasoning entirely

There is no effort level, including low, that skips the reasoning behind a decision entirely. Even at the lowest setting, the model still considers what it is actually looking at before acting on it.

What the levels genuinely shift is how much deliberation happens before a decision, not whether any deliberation happens at all; there is no mode that acts purely reflexively without looking first.

This is exactly why even a low-effort session still reads a file before editing it rather than guessing at its contents from the filename alone, the baseline discipline never actually goes away.

Framing low effort as "careless" is a common but inaccurate mental model; a more accurate one is "efficient for what it already understands well," spending less time specifically because there is genuinely less that needs working out for a well-scoped, familiar kind of request.

This is also why switching to low effort is not something to be nervous about for a task you are confident is simple; the floor under every level is the same baseline of actually looking before acting, not a sliding scale down to zero.