01 — The AI Crutch Relapse

Probability: 60% (highest of all failure modes).

This is the highest-probability failure in the entire pre-mortem, and it is not close. Rank it accordingly. If you find yourself relapsing here mid-roadmap, do not be surprised — be prepared. The relapse is the modal outcome; the disciplined exception is what we’re building.

Description

Three to four months into the roadmap, work at Zoho gets heavier. A deadline compresses. A new project lands on your plate. Your evenings shrink from 3 hours to 1 hour. You want to keep the C practice going — you told yourself you would. So you sit down at 10 PM, tired, with 45 minutes, and you have to choose: struggle for 45 minutes on a linked-list bug, or paste the failing code into Cursor and let it fix the problem in 60 seconds.

You paste it into Cursor. It works. You commit. You go to sleep.

And the next night, same choice. And the next. And by month 5, you have “kept up with the roadmap” — GitHub still shows commits, projects still ship — but you have not written a line of C from your own head in six weeks. The instincts you were building have quietly stopped forming. The compiler is now a black box to you again, just in a slightly different shape than before.

The scariest part is that from the outside, and from your own daily view, nothing looks wrong. The commits keep flowing. The rungs keep advancing. But the whole point of the roadmap — the reformation of your instincts — has silently stopped. You will discover this in month 8, in an study, when someone asks you to write a function on a whiteboard and you cannot.

Warning signals

  • Your commit messages get vaguer — “fix bug”, “update code”, “cleanup.” You are covering because you don’t remember what you actually changed.

  • You look at code from a week ago and don’t remember writing it. This is the loudest signal. Log it in your notebook.

  • You could not explain the function you just wrote to a friend without opening it.

  • The lab-notebook entries dry up. Because there’s nothing to predict when the AI writes the code — there’s no you-vs-machine dynamic to log.

  • You feel a small flinch of shame you can’t fully name. That flinch is data.

  • You have stopped using gdb. Because you never need it — the AI’s code compiles and works well enough.

Mitigation

Two levers, in order.

Lever 1: The “translate never, explain always” AI policy. Codified in 11_tools_setup/08_ai_assist_for_c.md. The rule:

  • ✅ Allowed: “Explain this K&R paragraph to me.” “What does this compiler warning mean?” “Why does this segfault?” “What are the trade-offs of strncpy vs strlcpy?”

  • ❌ Forbidden: “Write me a hash table.” “Fix this bug.” “Implement strtok.” “Give me a memory allocator.”

The line is: AI as tutor, never as ghostwriter. If you would learn something from writing it yourself, write it yourself. If you would only learn something by reading an explanation, ask for the explanation and then write the code yourself.

Lever 2: The monthly AI-free week. One week per month, chosen at the start of the month, Cursor uninstalled from the machine and Copilot disabled in your editor. Just man pages, K&R, gdb, and your own brain. Log what you build in that week. That’s your baseline. If AI-free weeks consistently produce meaningful work, your instincts are forming. If they produce nothing, your instincts are stalling, and the roadmap needs adjustment before the relapse gets worse.

Escalation trigger

If two consecutive AI-free weeks produce nothing shippable — the roadmap has broken your instincts. This is a real emergency, not a bad week.

Escalation protocol:

  1. Stop all AI tools for a full 30 days, not a week. Uninstall them from the machine.

  2. Go back to Rung 2 or Rung 3 material and rebuild the small things from your own head.

  3. Reset the calendar — you are not “behind”; you are relearning something that was hollowed out.

  4. Apply the full reset protocol from 09_summary_and_reset_protocol.md.

  5. Do not resume AI usage until you have shipped 3 consecutive weekly deliverables without any AI assistance.

This is expensive. It is much cheaper than arriving at Month 13 hollow and realizing the whole year was theatrical.

Return to README.md · Next: 02_the_zoho_promotion_glass_ceiling.md