AI-Driven Diet (68): Practicing "Habit Stacking" Supported by AI

[DAY 68] Tuesday, July 1, 2025

Introduction

It is now day 68 of the AI-driven diet. Recent meta-analyses report that the median time for health-related behaviors to enter the automated stage—what we call a "habit"—is 59 to 66 days. Today, just past that "critical point," I am reflecting on how to solidify dieting as a "habit." Following Heidi Grant Halvorson's "9 Things Successful People Do Differently," I am scientifically examining my own behavior and documenting the reality of how far I have been able to put my actions on autopilot with AI support.

1. Records

• Weight -2.2 kg compared to DAY 1

• Meals

  • Breakfast: Tofu, natto, corn

  • Lunch: Cashew nuts only

  • Snack: Strawberry shaved ice

  • Dinner: Tofu

  • Late-night snack: —

• Exercise Approx. 2,000 steps of walking, bodyweight strength training (100 reps each of squats, abs, and push-ups)

• Sleep Intermittent, but total amount is sufficient. Drowsiness during the day

• Physical Condition/Mood Mood fluctuated between 2 and 5 all day. Reached 5 after focused tasks

2. Retrospective (KPT Method)

Keep – Things to continue

1. Habit formation calendar and AI reminders

   1. Now that I have passed 66 days, the morning sequence of 'tofu + natto' -> '10-minute walk' -> 'record meal in AI app' has become semi-automated, 'like brushing my teeth.' This aligns with the meta-analysis indicator that 'automation occurs in about 2 months.'

2. 100 reps x 3 types of bodyweight training

   1. At the start of the month, 50 reps was my limit, but I reached 100 by using the 'slight increase method' of adding one rep every day. The load curve is gentle, but the ability to maintain progressive overload is a plus.

3. Quantifying mood

   1. By visualizing my mood score, I broke the old negative cycle of 'mood 2 -> stop activity,' allowing me to stick to a 'minimum walk' even when my mood was low.

Problem

1. Nutritional imbalance

   1. Nut-based lunches and tofu-only dinners—the nutritional value is monotonous, posing a risk of micronutrient deficiency.

2. Decrease in step count

   1. A sharp drop from 7,000 steps yesterday to 2,000 steps today. A decrease in NEAT is a common factor in weight loss plateaus.

3. Daytime sleepiness

   1. Even if the total amount of sleep is sufficient due to fragmented sleep, the proportion of deep sleep may be low. A pattern of sleepiness → craving for sweets → shaved ice occurs.

Try – Improvement Measures

1. Fixing walks with 'Habit Stacking'

   1. Recent reviews show data that 'linking a new habit immediately after an existing one triples the retention rate.' Therefore, I will add '5 minutes of reading' to the breakfast → walk routine to naturally extend the walking distance.

2. Protein + Colored Vegetable Rule

   1. Voice input to AI for every meal: '20g or more of protein + 1 dish of colored vegetables' → immediate feedback. Changed settings so the app issues a warning when protein intake is insufficient.

3. Scoring sleep quality

   1. Measure the proportion of deep sleep with a smart ring, and if it is less than 60%, limit caffeine intake the next day to before 2:00 PM.

4. 'Minimum 5,000 steps' alert

   1. Built an IFTTT integration where the AI 'suggests an additional walk before dinner' when the step count falls below 4,000 steps.

5. Mood 2 → Reset Protocol

   1. Following Halvorson's 4th habit, "if-then planning," I set a rule to immediately execute "30 seconds of deep breathing + stretching + 250 mL of water" if my mood is ≤ 2.

3. Today's Theme — Habitualizing Diet

The key to habit formation is "automation + predictable rewards." Now that I have passed 66 days, the main points for cementing these actions are the following three.

1. Decide on an "anchor" for the action

   1. "Habit stacking," which involves inserting a new habit (e.g., 10 squats) after an existing one (e.g., brushing teeth), is scientifically proven to be highly effective.

2. Setting immediate small rewards

   1. Receive a positive notification from the AI immediately after the action to provide a dopamine reward. The brain finds it easier to form habits when the distance between "action" and "reward" is short.

3. Visualization of progress and social proof

   1. Smartwatch badge features and SNS sharing serve as external accountability, with reports suggesting they increase continuation rates by 64%.

By mapping these to the framework of Halvorson's 9 habits (1. Concretization, 2. Progress Management, 3. if-then, etc.) and linking them with an AI action suggestion engine, the "calculation of habits" progresses.

Conclusion

While my actions have begun to automate after over two months, a decline in NEAT and nutritional imbalances remain challenges. Starting tomorrow, I plan to introduce "if steps are under 5,000, add a reading walk" and a "colorful vegetable challenge," feeding the data into the AI to receive feedback. For this "habit-driven" diet using behavioral science × AI, the next milestone is 90 days. I look forward to continuing this journey while enjoying the chain reaction of new habits.

English Version

Introduction

Day 68 (1 July 2025). According to recent meta-analyses, health-related habits become automatic after a median of 59–66 days. Having crossed that threshold, I reflect on how to lock my AI-driven diet into true habit. Grounded in Heidi Grant Halvorson’s “9 Things Successful People Do Differently,” I evaluate my choices, plans, and strategies.

1. Log

• Weight −2.2 kg (vs. Day 1)

• Food Breakfast: tofu, natto, corn / Lunch: cashew nuts / Snack: strawberry shaved ice / Dinner: tofu

• Exercise 2,000 steps + 100 reps each of squats, sit-ups, push-ups

• Sleep Fragmented but ample; daytime drowsiness

• Mood Oscillated between 2 and 5 all day

2. Retrospective (KPT)

Keep

• Calendar + AI reminders turned breakfast→walk→log into an almost automatic chain.

• 100-rep calisthenics via micro-progression.

• Mood tracking prevents low-mood shutdowns.

Problem

• Nutrient monotony (nuts & tofu).

• Steps dropped from 7 k to 2 k.

• Daytime drowsiness signals low deep-sleep ratio.

Try

1. Habit stacking to extend walks. Recent data show stacking triples adherence.

2. Protein 20 g + colored veggies rule with AI alerts.

3. Deep-sleep score via smart ring; caffeine cut-off if <60 %.

4. 5 k-step minimum alert with IFTTT.

5. Mood ≤2 protocol (deep breathing + stretch + water).

3. Topic — Making Diet a Habit

1. Anchor behaviors: insert new actions right after existing ones.

2. Instant micro-rewards: AI praise within seconds.

3. Progress visibility & social proof: badges and sharing raise adherence by 64 %.

Conclusion

My routines are becoming automated, yet NEAT and micronutrient balance need work. From tomorrow I’ll enforce a 5 k-step safety net and a “colorful-veggie challenge,” feeding the data back to AI. The next milestone is Day 90; I’m eager to see how these habit chains compound.