Focus isn't a discipline problem.
It's a restart-cost engineering problem.
In the next few minutes you will learn why a two-minute interruption can cost twenty-three minutes of real output, why a specific if-then plan beats a general intention at the exact moment willpower is weakest, how an AI system can hold a task's state so a restart costs seconds instead of minutes, and how to make that system resize itself from what actually happened last time instead of running a fixed checklist.
2. The Pit Crew Method
One F1 pit crew, mapped once onto each concept: the crew's entire job is making a stop cost seconds instead of a lost race.
Attention Residue
The Human Metaphor: A driver who pits without a crew has to check tire wear, fuel, and damage himself before re-entering, and every second of that check is a second not spent racing. Attention residue works the same way: part of the mind keeps re-checking the task it left, even after moving to the next one.
- Unit: minutes of degraded output per switch
- Fixes Failure: pure willpower and distraction-blocking, which target the interruption instead of the restart
- Real-world Parallel: Leroy (2009), attention residue
The If-Then Trigger
The Human Metaphor: The crew chief never asks the driver to decide when to pit. The radio call, "box now," is a pre-agreed rule that fires the moment a condition is met, so nothing has to be decided under pressure. An implementation intention is that same rule, written for a workday.
- Unit: one pre-committed if-then rule
- Fixes Failure: vague intentions like "I will focus more"
- Real-world Parallel: Gollwitzer (1999), implementation intentions
AI State Capture
The Human Metaphor: The pit board never asks the driver to remember tire pressure or the last lap's damage report. It hands over one readout: what changed, and what to do next. An AI system that holds a task's state does the same job for stalled work.
- Unit: one restated task state per session
- Fixes Failure: relying on memory or a stale to-do list to rebuild context
- Real-world Parallel: an on-call runbook handoff between shifts
Adaptive Sizing
The Human Metaphor: A good crew does not run the same twelve-step checklist on every stop regardless of what happened last time. They read the telemetry and adjust the next stop, shorter or longer as needed. The loop should size the next micro-task the same way.
- Unit: one resized micro-task per loop
- Fixes Failure: fixed daily goals and streak counters
- Real-world Parallel: adaptive difficulty in structured training programs
3. Attention Residue
THE DIAGNOSISWhat actually gets destroyed when a task gets interrupted.
The Driver Who Pits Alone
Mara is deep in a budget model when a meeting invite pulls her onto a call. She comes back forty minutes later, opens the spreadsheet, and spends the next eighteen minutes re-reading formulas she wrote that same morning before she trusts herself to keep going.
The forty-minute meeting is not what cost her the morning. Organizational psychologist Sophie Leroy found that people who switch tasks without finishing the first one carry part of their attention into the next task, a residue that gets worse the less resolved the interruption felt. Mara's spreadsheet did not close cleanly in her head, so her brain kept working it in the background, at the expense of the meeting and then the return to the spreadsheet.
THE FIX: MEASURE THE RESTART, NOT THE INTERRUPTION
Gloria Mark's research puts the average recovery at about 23 minutes. That number is not fixed. It shrinks sharply once something outside the person's head already holds the task's state, because the brain has nothing left to reconstruct.
The 1-Sentence Recall Lock-In
"The interruption is free. The reconstruction is what you're actually paying for."
Visual Memory Anchor 1: The Residue Curve
// crude model: residue scales with how unresolved the task felt at handoff
def restart_cost_minutes(unresolved_fraction, held_state=False):
base = 23 # Mark (2023), average full-recovery minutes
if held_state:
return round(base * 0.02 * unresolved_fraction, 1) # AI restates state on resume
return round(base * unresolved_fraction, 1)
# Mara's spreadsheet, left 70% unresolved
restart_cost_minutes(0.7) # -> 16.1
restart_cost_minutes(0.7, held_state=True) # -> 0.34. The If-Then Trigger
THE TRIGGERRemoving the decision at the exact moment willpower is weakest.
The Radio Call: Box Now
Dev tells himself every Monday that this week he will finally focus on the roadmap doc. By Wednesday it is still untouched, not because he does not care, but because "focus more" never specifies when, where, or what the first move actually is.
Psychologist Peter Gollwitzer's research on implementation intentions found that a plan phrased as "when X happens, I will do Y" produced far higher follow-through than a general intention to act. The plan does the deciding in advance, so nothing has to be decided in the moment discipline is at its lowest.
THE FIX: WRITE THE IF-THEN, NOT THE GOAL
"I will focus on the roadmap" is a goal. "When my 9am calendar block starts, I open the roadmap doc and read the AI's last-session note" is an implementation intention: a specific cue tied to a specific first action.
Anti-Pattern: The Vague Intention
An intention without a specific trigger rarely survives contact with a real calendar. A rule that says merely "focus more" competes with every open tab for attention, and loses, because it never names the moment it should fire.
"If the plan does not name a cue and a first action, it is a wish, not a trigger."
Visual Memory Anchor 2: The If-Then Chain
triggers:
- cue: "9:00 calendar block starts"
action: "open roadmap.md, read last_state note"
max_start_delay_sec: 120
- cue: "AI session-end note flags 2 incomplete blocks"
action: "shrink next micro-task by half"5. AI State Capture
THE HANDOFFWhat an AI can hold that a to-do list cannot.
The Pit Board
Sam closes a half-written incident report on Friday, and re-opens it Monday with no memory of which two paragraphs were already fact-checked and which were still guesses.
A to-do list item, "finish incident report," carries none of that. It names the task, not its state. Sam ends up re-reading the whole document to rebuild a model he already had three days earlier, paying the attention-residue cost from section 3 a second time.
THE FIX: ONE RESTATE, NOT A RE-READ
An AI system that can read the document and the last session's note can restate, in three lines, what's done, what's blocked, and the next concrete step. That restatement replaces the re-read.
The 1-Sentence Recall Lock-In
"A task name tells you what to do. A state note tells you where you actually are."
Visual Memory Anchor 3: State Survives the Gap
// paste at the top of a fresh chat
SESSION START
Task: [one line]
Last state: [paste the AI's last session-end note]
-> exactly one next action, under 2 minutes to start
// paste before closing the chat
SESSION END
-> done / blocked / next, in three lines
-> flag if the last two sessions ended incomplete6. Adaptive Sizing
THE SELF-CORRECTIONWhy the next step should be sized from data, not from a plan made yesterday.
Telemetry, Not a Checklist
A habit app gives everyone the same daily target: write for 25 minutes. Priya hits it three days running, then misses four in a row and stops opening the app altogether.
The app never asked why the streak broke. It could not, because a fixed daily target cannot read what happened in the sessions that failed. It only knows whether Priya showed up, not whether the step she was given fit the time and energy she actually had.
THE FIX: SIZE THE NEXT STEP FROM THE LAST ONE
An AI that reads the session-end note has a real completion signal to work with. Two unfinished blocks in a row should shrink the next micro-task. Two blocks finished early should grow it. The target adapts to the person instead of the reverse.
Anti-Pattern: Optimizing for the Streak
Gamified focus apps reward showing up, not finishing, so the metric that goes up (days in a row) can be disconnected entirely from the metric that matters (whether the real task moved forward).
"A system that cannot tell 'showed up' from 'finished' will optimize for the wrong one."
Visual Memory Anchor 4: The Sizing Branch
{
"block_id": 214,
"task": "roadmap doc: risks section",
"planned_minutes": 25,
"actual_minutes": 40,
"completed": false,
"next_action_size": "shrink"
}7. The One Rule Underneath the Loop
DISTILLED RULEHow diagnosis, trigger, handoff, and adaptation collapse into one thing to measure.
Optimize restart cost, not raw distraction count
Attention residue names what breaks. The if-then trigger removes the decision cost of starting again. AI state capture removes the reconstruction cost. Adaptive sizing keeps the next step realistic instead of aspirational. None of the four pieces matters much alone. Together they turn each interruption from an hour-long tax into a twenty-second one, which is the only number worth putting on a dashboard.
FROM LOG TO RULE
session_end_note:
done: string
blocked: string
next_step: string
minutes_planned: int
minutes_actual: int
completed: boolFour fields, filled once per session, are the entire data set the adaptive loop needs.
8. Reader Retention & Active Recall System
LONG-TERM MEMORY LOCKInteractive self-testing tools designed to lock these mental models into your long-term memory before you leave.
Why does a short interruption cost 23 minutes instead of the length of the interruption itself?
What makes an implementation intention stronger than a goal like "I will focus more"?
What does an AI-held state note actually replace?
Why can't a streak counter size the next micro-task correctly?
What single number should a team actually track to know if this system works?
9. The Unified Loop
Four concepts, one system: what changes when they run together instead of alone.
Willpower-Only Focus
Each interruption is absorbed by discipline alone. The interrupted task's state lives only in one person's head, so every resume is a full reconstruction, and the daily target stays the same regardless of how yesterday actually went.
Roughly 23 minutes of degraded output per switch (Mark, 2023), paid again every time, with no mechanism to get cheaper over time.
The AI-Held Focus Loop
A trigger starts the session, an AI restates the task's state instead of the person reconstructing it, and the next micro-task is sized from what actually happened last time.
A restart cost that trends toward twenty seconds instead of twenty-three minutes, and a system that gets better calibrated the longer it runs.
Visual Memory Anchor: The Full Loop, One Trigger to the Next
10. Build Order
A pragmatic adoption order that avoids shipping a system nobody feeds with real data.
Write the two prompts
- Session-start and session-end prompts
- Works with any AI assistant already in use
- Relies on remembering to paste the last note
Standardize the session-end note
- One line: done / blocked / next
- Makes the AI's restate possible at all
- Needs a team norm, not just a personal habit
Let sizing run on real data
- Next micro-task shrinks or grows from actual completions
- Replaces streaks with a real signal
- Needs 2-3 sessions of history before it can size well