The fastest way to ship code right now is to describe a problem to an AI and paste the answer. The fastest way to forget how to solve problems is to do it every time.
AI is not the enemy. The enemy is the reflex.
The quiet trade behind every prompt
Ask AI to explain the error. Ask AI to write the function. Ask AI to name the variable and review the PR. Each request saves minutes. Together, they can remove the part of the work that used to build judgment.
Cognitive surrender is what happens when we hand over more than the typing. We hand over the model of the problem: the attempt to decompose it, the edge case we almost missed, the bug we had to trace.
It feels efficient because the output is usually fine. That's exactly why it's easy to miss the cost. Fine output hides the missing rep.
Why 15 minutes is enough
Skill comes from retrieval and struggle, not from watching answers appear. You need a loop where you read the problem, form a plan, write code, and test it — without outsourcing the first three steps.
Fifteen minutes is short enough to do daily and long enough to complete that loop. Even on an easy kata.
A protocol that doesn't cheat
- Read the problem twice. Write the input, output, and one edge case in plain words.
- Draft the steps before touching the editor.
- Set a 15-minute timer. Solve without AI.
- If you get stuck, write down exactly where the plan breaks.
- Only then compare with the solution or ask AI.
- Save one takeaway: the trick, the edge case, or the bug.
The rule: no prompts until you have a plan. The comparison is where AI is most useful — after you have something to compare.
8 free katas to start (10 minutes each)
All of these are free and available in JavaScript, TypeScript, and Python.
Strings
- Valid Palindrome — normalize the input, then compare from both ends.
- Is Anagram — count letters and compare frequencies.
- Max Distance — parse movement commands and decide when
*helps.
Arrays
- Merge Intervals — sort by start, then merge overlaps.
- Find First Repeated Element — track seen values in one pass.
- Unique Sorted — dedupe, then order.
Objects and recursion
- Organize by Category — group with reduce and sum quantities.
- Tree Height — one recursive call per branch.
When 10 minutes feels easy
Add one of these:
- Flatten Array — recursion over nested data.
- Decode Parentheses — stack-based parsing.
- Max Non-Adjacent Sum — dynamic programming with memoization.
The point is not the algorithm
You are not trying to become a competitive programmer. You are trying to keep the loop: read, reason, write, verify.
AI amplifies whatever you bring to it. If you bring a sharp model of the problem, it amplifies that. If you bring nothing, it replaces that too.
Fifteen minutes a day is a small price to stay in the loop.
