Habits are input-output mappings that have been practised so many times they fire automatically. The context (input) directly activates the response (output) without needing deliberation. You walk into the kitchen, your hand reaches for the cupboard. You sit at your desk, your thumb opens the social app. The mapping is so strong that the output often starts before you've even registered the input.

So what can you do? Change the input. If the context cues the habit, alter the context: rearrange the furniture, change your route, remove the trigger entirely. Or add a new input at the boundary: put the phone in another room, put a note on the cupboard, wear a rubber band on your wrist as a physical reminder. You're not fighting the mapping head-on; you're disrupting the cue so the mapping doesn't fire in the first place, or inserting a new first step that redirects the sequence.

Your brain builds a predictive model of what you do in each context. When you've done something hundreds of times in a particular situation, the model becomes very confident: "In this context, I do this thing." That confident prediction drives action automatically. The habit isn't just a mapping; it's an expectation, and the brain acts to fulfil its own predictions.

So what can you do? Retrain the prediction. Practise the new behaviour in the exact context where the old habit used to run, and do it enough times that the new prediction becomes confident. Expect this to take time---the old prediction has infrastructure behind it. You can also adjust precision by making the new behaviour more salient (visual cues, social accountability) so the brain weights it more heavily when generating predictions.

Habits win because control is expensive and habits are cheap. By the time you notice the habit starting, it's often already finished. Control is designed to stay out of the way unless there's conflict or accountability, so well-trained routines execute before control even wakes up.

So what can you do? Don't rely on runtime willpower. Design the situation so control doesn't have to fight: preload your intention before the cue (set an alarm, tell someone your plan), add friction to the old habit (put the snack in a locked cupboard), or build accountability into the moment (work in a shared space where people can see you). You're setting up the environment so the habit has to compete with something other than your willpower.

Habits are chunked sequences: cue triggers chunk A, which triggers chunk B, which triggers chunk C. Once the first chunk fires, the rest follow like dominoes. This is why "just stopping" mid-habit feels so difficult---you're trying to interrupt a sequence that's designed to run fluently.

So what can you do? Work at the boundaries between chunks. Identify where one sub-routine hands off to the next, and insert a new chunk there---a pause, a breath, a different physical action. Practise that interruption in context so it becomes part of the sequence. You're not trying to delete the habit; you're rewiring the sequence so it assembles differently.

Habits are extreme bias: fast, consistent, efficient, and inflexible. The pathway is so consolidated that alternatives don't even get sampled. This is great when the habit serves you, but it means changing the habit requires introducing noise---loosening the pathway, trying alternatives, tolerating inconsistency while the new route builds infrastructure.

So what can you do? Expect variability during the transition. Breaking a habit means temporarily increasing noise (you'll be slower, less consistent, more effortful) until the new pathway consolidates. Use variability deliberately: practise the new behaviour in different contexts, at different times, with different cues, so it builds flexibility rather than just swapping one rigid habit for another. And don't mistake a slip-up for failure---noise is part of the process.

Habits aren't one monolithic thing. They're coalitions: one part proposes the familiar action ("check phone"), another part objects ("no, stay focused"), but the habitual part has so much infrastructure that it wins the vote before the objection even finishes loading. The sense that you "didn't mean to" is the losing coalition making itself heard after the fact.

So what can you do? Give the habitual part a better job. If the phone-checking part wants novelty or relief, design an alternative that serves that need without derailing your goal (a scheduled break, a different source of novelty). Practise the alternative until it has enough infrastructure to compete. You're not suppressing the part; you're negotiating with it and offering it a role that doesn't sabotage the whole coalition.