Pre-Mortem 01 — The AI Crutch¶
Failure Mode: Using AI as a solution generator instead of a concept clarifier.
Probability: HIGH — 80%
Peak Risk Period: M1 onward, continuously. This failure mode has no safe phase.
Why the Probability Is This High For You Specifically¶
You are an Applied ML Engineer at Zoho. You work with AI tools daily. Your professional muscle memory is to use AI to generate output efficiently. In your work context, this is correct — you should use AI for code generation, documentation, boilerplate, pipeline scaffolding. Efficiency is the goal.
In DSA learning, the exact same reflex — “let me just get the answer” — is harm, not efficiency. The struggle with the problem is the learning mechanism. The moment AI removes the struggle, it removes the thing that causes the neurons to wire. You are left with a solved problem and an unchanged brain.
This is not a willpower problem. It is a muscle memory problem. The reflex that makes you a productive engineer at Zoho is the reflex that will destroy this learning plan if you don’t consciously override it.
What It Looks Like¶
Week 1: You’re stuck on a graph problem. You ask AI to explain the approach. The explanation is good. You implement it yourself. This is fine — this is concept explanation, which is allowed.
Week 3: You’re stuck on a DP problem. You ask AI for the approach, then the implementation hints, then “can you show me a clean version.” You read the clean version, understand it, and mark the problem solved. This is the beginning of the crutch.
Month 2: You’re stuck on a problem. Without hesitation, you open AI and ask for the solution. You read it, understand it superficially, mark it solved. The session ends. You feel productive. You have learned nothing.
Month 4: You cannot solve a medium problem you haven’t seen before. You don’t realize why. The AI has been scaffolding your understanding and you’ve mistaken the scaffold for the building.
The degradation is gradual and invisible in the moment. You will not notice it happening. The early warning signs are the only way to catch it.
Early Warning Signals¶
Check these weekly. If any are true — invoke the mitigation immediately.
You haven’t solved a problem without AI assistance in 3+ consecutive sessions
You are moving through problems “fast” but cannot explain why any solution works
You feel reluctant to attempt a problem before consulting AI
Your editorial dependency rate (from
03_benchmark_hygiene.md) is above 80%When you try to solve a problem without AI, you feel unusually stuck — more stuck than you remember being in earlier sessions
The Mechanism (Why AI Assistance = Harm in This Context)¶
Your ML background gives you the exact model to understand this. In supervised learning, if you give the model the labels before training (label leakage), it fits to the labels directly and learns nothing generalizable. AI solution generation is label leakage. The “model” (your brain) is fitting to the provided answer rather than learning the underlying pattern. The result is zero-shot generalization failure — you can’t solve problems you haven’t seen the answer to.
The only way to develop pattern recognition is to struggle with patterns. The struggle is not inefficiency — it is the training signal.
Concrete Mitigation¶
Rule 1: AI is allowed for concept explanation. Never for solution generation.
Allowed |
Not Allowed |
|---|---|
“Explain what a segment tree is and why it’s used” |
“Solve this segment tree problem” |
“Why does BFS guarantee shortest path in an unweighted graph?” |
“What’s the approach for this BFS problem?” |
“What does it mean for an algorithm to be O(V+E)?” |
“Optimize my solution for this graph problem” |
“Explain the difference between memoization and tabulation” |
“Show me the DP solution for this problem” |
If you’re unsure which side of the line a query falls on: if answering the question would reduce the amount of thinking you need to do on the current problem — it’s not allowed.
Rule 2: Use a separate browser profile or device for DSA practice with no AI access.
This is not about trust — it is about removing the frictionless path. If using AI requires switching browsers and logging in, you will do it less. The reflex needs a speed bump, not a wall.
Rule 3: Honor system accountability.
If you have an accountability partner, tell them about this rule. If you don’t, your lab notebook is your accountability. Record honestly: “AI used for solution today” is a log entry, not a confession. But you have to write it down.
Escalation Trigger¶
If you’ve used AI for solution generation 3+ times in one week:
Mandatory 2-week AI-free sprint. No AI access during DSA practice sessions, for any reason. Use editorials from official sources (LeetCode editorial, Codeforces editorial) if you’re stuck — these are allowed because they require more cognitive effort to parse than AI output, and they teach the problem-solving vocabulary of the community.
Log the 2-week AI-free sprint in your sprint planning template. It is not punishment — it is recalibration.
The Meta-Point¶
You came to this plan because you felt that too much code was being churned out without actually learning. The AI crutch is the exact same mechanism, applied to DSA. If you allow it here, you will end 9 months with a solved problem count and a brain that still can’t solve an unseen problem independently.
The point is not the problems solved. The point is the brain that gets built. Protect the training signal.