F1 — AI-Crutch Relapse

Probability without countermeasures: 70%. Window: M1-M3, with echoes possible through M6. Severity if unchecked: catastrophic — the entire plan becomes theater.

What This Failure Mode Actually Is

You open your editor to write a LinkedList from scratch as part of the 50-programs challenge. You know how it should work — nodes, pointers, insert-at-head, insert-at-tail. But your fingers do not know. There is a five-second silence where you stare at the cursor.

And then the reflex kicks in: Cmd+K, “generate a doubly-linked list in Java 21.” Claude/Cursor produces a clean implementation. You read it, you understand it, you commit it. You feel productive. You mark the exercise done.

You learned nothing. The muscle that was supposed to be rebuilt got skipped again.

This is the failure mode that this entire roadmap exists to defend against. If you fall into it, none of the other 12 phases matter. You will finish August 2027 with 13 months of AI-assisted work and the same two-year gap in your hand-coding.

Why the Probability Is 70%

You have been outsourcing Java thinking to AI agents for approximately two years. That is not a habit — that is a reflex circuit. It runs faster than your conscious intent. When your fingers hit a wall, they reach for the agent before your brain has time to say “wait, I am supposed to be practicing.”

The 70% number is not pessimism. It is a calibrated estimate for engineers with your exact profile: (a) AI-fluent daily user, (b) full-time job that rewards speed over depth of understanding, (c) prior Java competence that makes you think you can “just read the AI output and remember.” You cannot. Reading is not writing.

Early Warning Signs

Catch these before they compound. Any two in one week = read the mitigation section same day.

  • You typed “explain this Java code” to Claude in Month 2. (You are supposed to be writing the Java code, not explaining it.)

  • You skipped a 50-programs drill because “I already know how this works.”

  • You opened Cursor’s inline suggest during a Phase 01 or Phase 02 exercise.

  • Your “checking your work” is actually rewriting your work.

  • You finished a drill in 8 minutes when the estimate was 30-45.

  • Your git log shows commits with no wrong versions before them. Real learning leaves failed drafts in the history.

  • You cannot explain your own code from yesterday without opening the file.

Root Causes

This is not a discipline problem. It is a wiring problem, and the wiring is deep:

  1. Speed addiction. AI agents made you fast at ML work. Slow feels like regression, even when slow is the point.

  2. Ambiguity aversion. Not knowing the answer feels bad. AI removes the discomfort in 3 seconds.

  3. Sunk-cost of prior competence. You used to write this stuff. Admitting you cannot anymore feels like failure, so you shortcut around the admission.

  4. Job-context bleed. Your Zoho work uses AI agents all day. When you sit down at 8pm for Java practice, the same fingers do the same thing.

The defense has to address the wiring, not the willpower.

Mitigation — Non-Negotiable in M1-M2

These are not suggestions. They are the entire point of the plan.

Layer 1: Physical Separation (M1-M2)

  • Physical notebook for the 50-programs challenge. Yes, actual pen and paper. You write the algorithm by hand before you type it. The pen has no autocomplete. The pen forces the mind. Buy an A4 unruled notebook this week. Label it JAVA_LAB_2026.

  • Separate machine or user profile. Create a macOS user account called java-practice. No Cursor. No Copilot. No Claude desktop app. VS Code with vanilla settings only, or IntelliJ Community with AI Assistant disabled. This is your practice environment for M1-M3.

  • Browser profile lockdown. A separate Chrome/Firefox profile for practice sessions with no extension access to your AI accounts. Bookmark: docs.oracle.com, dev.java, the JDK Javadoc. Nothing else.

Layer 2: Time Fencing (M1-M6)

  • AI-agents-off block: 6-8pm every weekday. Block it in your calendar. Phone in another room. Slack DND. This is the sacred slot.

  • The 45-minute rule. For any Phase 01/02 exercise, you must sit for 45 minutes with pen and keyboard before you consult any external source — AI, Stack Overflow, or Baeldung. Set a timer.

  • The reference tier. When the 45 minutes ends and you still need help: first the JDK Javadoc, then a physical book (Effective Java, Java Concurrency in Practice), then Stack Overflow, and only then — last — an AI agent. And when the AI agent gives you the answer, you close the tab, wait 10 minutes, and re-type the solution from memory.

Layer 3: Verification (Monthly)

  • The hand-writing count. By the end of M2 you must have at least 30 hand-written programs in your JAVA_LAB_2026 notebook and their typed counterparts committed to a 50-programs repo.

  • The explain-your-own-code test. Once a week, open a file you wrote 5+ days ago and explain it out loud (or in a voice memo) without looking at comments. If you cannot, that program does not count — rewrite it.

Escalation Trigger — Read This Twice

If by the end of M2 (approximately end of August 2026) you have fewer than 30 hand-written programs, STOP EVERYTHING and reset.

Do not continue to Phase 02. Do not tell yourself you will “catch up in M3.” You will not. The wiring has not changed and it will get harder, not easier, once DSA and concurrency start.

The reset protocol is in 09_summary_and_reset_protocol.md. Follow it exactly. A reset in M2 costs you 2 weeks. A relapse discovered in M6 costs you the whole plan.

What Success Looks Like

By end of M3 you should be able to:

  • Write a HashMap from scratch (open addressing or chaining, your choice), by hand, in under 40 minutes, with no reference.

  • Explain, without looking, what volatile and synchronized actually do at the memory-model level.

  • Feel a positive discomfort when you reach for an AI agent during practice. The reflex is still there, but there is now a competing reflex — the memory of writing it yourself.

If you have that by end of M3, the crutch is broken. From M4 onward you can loosen the physical-separation rules and treat AI agents as a colleague you consult occasionally, not a colleague who does your work.

The Honest Note

The hardest week for this failure mode is Week 2, not Week 1. Week 1 you are motivated. Week 2 you are tired, the novelty is gone, the drills feel repetitive, and the reflex is loudest. Expect it. Plan for it. The 45-minute rule and the physical notebook are specifically designed for Week 2, not Week 1.

Every Java engineer who came before the AI era learned this way. You are not being asked to do something impossible — you are being asked to do the thing everyone did before it became optional.

Previous: README.md | Next: 02_fm_work_load_erosion.md