There is one mental model that changed more of my daily behavior than every productivity book combined, and it fits in a sentence:
Human attention is a single-threaded runtime with an extremely expensive context switch and no reliable serialization of state.
Everything else follows from that.
You are not running tasks in parallel
When you "multitask," nothing runs concurrently. The scheduler is interleaving: it suspends task A, tears down the context, loads task B, and — crucially — much of task A's working state is simply lost rather than saved. It has to be rebuilt from scratch when you return.
In software we would call this a catastrophically bad implementation. We would rewrite it. You cannot rewrite the runtime, so the only remaining lever is the schedule.
The three costs of a switch
- Teardown. Whatever you were holding in working memory decays as soon as attention leaves.
- Reload. Getting back to depth on a non-trivial problem takes real minutes, not seconds — and the deeper the problem, the longer the reload.
- Residue. The previous task keeps consuming capacity in the background. You are not fully on B for some time after you leave A, which is why a two-minute Slack check does not cost two minutes.
Costs one and two are visible. Cost three is the one that quietly determines whether your day produced anything.
What follows in practice
Batch by context, not by priority. All code review in one block, all writing in another. Priority ordering across mixed contexts maximizes switches — the single worst thing you can do to this runtime.
Make interruptions synchronous by exception. Default everything to async. A notification is a preemptive interrupt with no priority level attached, which means every message is implicitly claiming to be the highest-priority thing in your life. Almost none of them are.
Protect block boundaries more than block length. A 90-minute block that gets one interrupt is worth less than a clean 60-minute block. Guard the edges.
Stop at a resumable point, never a random one. Same principle as leaving a failing test or a half-written function as your bookmark: encode the reload state externally before you let attention leave.
The uncomfortable implication
If this model is right — and everything in my own time audits says it is — then most of what gets labeled a discipline problem is actually a scheduling problem.
You are not lazy. You are running a well-designed workload on a schedule that guarantees thrashing, and then blaming the workload.
Fix the schedule first. Then find out whether you had a discipline problem at all.