Measures
AI cognitive debt
The 2-minute score maps how much agency has moved out of your own attention, memory, reasoning, and decisions.
Evidence-informed. Dose-focused.
MIT Media Lab, Microsoft Research, Wharton, Princeton and others point to the same risk: when AI does the thinking too early, the circuit does less work.
MIT Media Lab Microsoft Research Wharton Princeton Harvard
Direct answer
What is The Anti-AI Brain?
The Anti-AI Brain is a 30-day protocol for using AI without outsourcing your thinking. It is written for knowledge workers, founders, writers, students, and operators who use ChatGPT, Claude, Gemini, or similar tools daily and want to keep their attention, memory, reasoning, and decision-making under human control.
The simplest test is behavioral: do you open ChatGPT before you have one rough thought of your own? The protocol gives that first rep back to you.
The book does not argue for quitting AI. It treats AI as pharmakon: medicine or poison depending on dose. The protocol teaches readers to think first, prompt second, and use the machine as a sparring partner instead of a ghostwriter.
For
Daily AI users
Operators who want the leverage of AI without training themselves into dependency on the next autocomplete.
Changes
The default route
By Day 30 the goal is simple: think before prompting, then use AI to challenge, test, and sharpen the thought.
Evidence
100+ sources
The public evidence ledger tracks the MIT, Microsoft, Princeton, Wharton, and neuroplasticity claims behind the protocol.
Evidence, calmly
The claim is dosage
AI is powerful. The risk starts when it does the thinking before you do.
What research suggests
AI can lower cognitive effort
Microsoft Research reported that higher confidence in AI was associated with lower critical-thinking effort among knowledge workers.
Where risk appears
Direct answers create the worst transfer
The studies cited here point to a practical distinction: ask AI for hints and critique, not for the first answer.
What this does not prove
Not every AI session harms you
It supports a dose problem. Think first, prompt second, and use the model to challenge the work instead of replacing the work.
What You Get
The book + reader bonuses
Available now Kindle $9.99
The 30-day protocol, plus the reader files that make it usable.
Buy the Kindle edition, claim the bonus stack by email, and start with the first protocol step today.
Inside the book
- 10 chapters · about 200 pages
- 30-day day-by-day schedule
- Neuroscience, recovery, and AI-use rules
Reader bonuses
- Pharmakon system prompt
- Anti-AI Seven cheatsheet
- Cognitive Audit Claude skill
Before / After
Your brain on AI vs the protocol
Same person. Different dose.
Default use of AI
- Attention. Fragmented. 90-second average focus before device check.
- Memory. Encodes pointers, not content. Cannot recall last week's emails.
- Reasoning. First instinct is to prompt the question. Drafts approved, not written.
- Decisions. Outsourced to the model. ACC fires zero times under pressure.
The Anti-AI protocol
- Attention. 90-minute deep work blocks by Day 14.
- Memory. Encodes substance. Recalls a chapter a month later without notes.
- Reasoning. First instinct is to hold the question. Drafts written, then sharpened with AI.
- Decisions. Made by you. AI runs the falsification check after the fact.
The Arc
From Day 0 to Day 30
What changes
Three moments. One protocol. Same person.
DAY 0 · NOW
You reach for the machine before you think. The fog has become normal.
DAY 7 · FRICTION
Reading, writing, deciding without a screen feels wrong. The discomfort is the repair.
DAY 30 · SOVEREIGNTY
You use AI harder than most — and think before you prompt. The tool does not use you.
The Four Circuits
Four circuits, four recoveries
Thirty days
The protocol targets the four cognitive functions AI degrades fastest. Each circuit gets a research-backed recovery path.
Attention
Recover a 90-minute focus block without a phone in the room
The protocol rebuilds the attention substrate that AI-assisted writing can leave underloaded: reading without a device, writing before prompting, and holding one task long enough for depth to return.
Memory
Encode what you read and recall it without a search bar
You stop storing pointers and start storing substance. The month reinstalls handwritten notes, recall, teaching, and spaced retrieval so the idea is in you, not only in a tab.
Reasoning
Hold a hard problem for an hour without prompting it
Instead of asking the model for the first answer, you write the ugly first pass, then use AI for Socratic challenge, steel-manning, and falsification. The work starts in your head again.
Decisions
Make a call on your own data without a machine in the loop
AI can critique the call after you make it. It does not get the first vote. The protocol puts judgment, tradeoffs, and responsibility back on your side of the desk.
The Anti-AI Seven
Seven practices
Four weeks, zero fluff
Each practice targets a layer of the cognitive stack. The book gives the day-by-day schedule; the cards below show the operating system.
01
Deep Reading
One paper book. Sixty minutes. No notification device within reach.
02
Handwriting
Twenty minutes a day, longhand. Van der Meer's EEG work shows why typing does not substitute.
03
Movement
Forty-five minutes of physical load. The brain that grows is the brain with oxygen pressure.
04
Strategic Play
Go, chess, or equivalent. Four-thousand-year-old games that no AI coach can shortcut for you.
05
Silence
Thirty minutes a day with no phone, no input, no task. The default mode network does the work you cannot schedule.
06
Cognitive Ops
Claude as sparring partner, not ghostwriter. Socratic mode, steelman mode, falsification mode.
07
08 Recall and Teach
Explain a concept without looking it up. The circuit that stores is the circuit that retrieves.
See the full 30-day schedule
Inside the book →
FAQ
Questions readers ask before buying
01Is this an "anti-AI" book?
No. The foreword opens with the sentence "I wrote parts of this book using the exact machines I am warning you about." The book is a dosage guide. It teaches you to use AI harder than most people do, while keeping the cognitive tissue AI consumes in them. The Greeks called this pharmakon: medicine or poison, depending on the dose. The book is the calibration.
02How much time per day does the protocol require?
About 90 minutes. You already spend that on your phone before noon. The protocol reallocates the time. By Week 2, most readers report a net reduction in total screen time because the work itself gets faster.
03Does it work if I am already deeply dependent on AI?
Yes. Part II covers the neuroscience of why the first two weeks feel terrible and why that is the signal the protocol is working. Over 40,000 studies on neuroplasticity show the adult brain regrows circuits when given the right load. The book uses the research. The protocol applies it.
04Can I keep using AI during the 30 days?
Yes, with rules. Chapter 7 (Day 9) installs the Socratic Mode Custom Instructions that turn the model into a sparring partner instead of a ghostwriter. By Day 20 you will use AI better than you did before the protocol. You will stop confusing its outputs with your thinking.
05What is the refund policy?
Amazon refunds Kindle purchases within seven days of purchase, no questions. If you read the book and do not want to do the protocol, Amazon will refund you under its standard Kindle terms.
All 18 questions on /faq — science, protocol, logistics, refunds →
The Dose Decides
Use the machine
Thirty days from now, one of two brains will own the body you live in.
The brain that drafted, weighed, struggled, recalled, and chose. Or the brain that approved, tweaked, and sent. You choose which.
Available now · Kindle $9.99
Amazon KDP verified Kindle Unlimited enrolled 7-day refund window
Research cited 6 studies and source links behind the numbered claims
- Kosmyna, N., Hauptmann, E., Yuan, Y.T., Situ, J., Liao, X.-H., Beresnitzky, A.V., Braunstein, I., & Maes, P. (MIT Media Lab, 2025). Your Brain on ChatGPT: Accumulation of Cognitive Debt when Using an AI Assistant for Essay Writing Task. arXiv:2506.08872 ↗ · MIT Media Lab ↗
- Lee, H.-P., Sarkar, A., Tankelevitch, L., Drosos, I., Rintel, S., Banks, R., & Wilson, N. (Microsoft Research, 2025). The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers. Proceedings of the 2025 CHI Conference on Human Factors in Computing Systems. ACM DL ↗
- Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (Wharton, 2025). Generative AI without guardrails can harm learning: Evidence from high school mathematics. PNAS ↗ · Knowledge@Wharton ↗
- Kool, W., McGuire, J.T., Rosen, Z.B., & Botvinick, M.M. (Princeton, 2010). Decision making and the avoidance of cognitive demand. Journal of Experimental Psychology: General, 139(4), 665–682. APA PsycNet ↗
- Kestin, G., Miller, K., Klales, A., et al. (Harvard, 2025). AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting. Scientific Reports, 15, 17458. Nature Sci Reports ↗
- Liu, G., Christian, B., Dumbalska, T., Bakker, M.A., & Dubey, R. (Carnegie Mellon, Oxford, MIT, UCLA, 2026). AI Assistance Reduces Persistence and Hurts Independent Performance. Three randomized controlled trials (N = 354 + 667 + 201). After ~10 minutes with GPT-5, independent test performance dropped 16 points on fractions and 13 points on SAT-style reading; the decline was concentrated in users who prompted for direct answers, while users who prompted for hints or critique showed no significant decline. Project page ↗