How to Use AI Without Losing Your Skills

I use AI every day. I also still practise my craft without it, regularly, on purpose. These are not a contradiction — if you draw a line between what you want to automate and what you want to master, they stop being in tension.

This is the exact framework I live by. It has survived two years, a career shift, and three different AI-tool cycles. It still works. Here it is.

The 50/50 rule

Half your work, you do yourself, slowly, to learn. Half, you let AI assist, to ship fast. The split is not literal — some weeks are 70/30, some are 30/70 — but the principle holds: there is a "do it yourself" lane and an "automate it" lane, and you are clear-eyed about which is which.

• Half your work you do yourself, slowly, to learn
• Half you let AI assist, to ship fast
• Never let the "AI half" grow until the foundational skill is solid
• Once a month, do a task entirely without AI and notice the gap

Deciding which lane a task goes in

Here is the decision tree I use. It takes ten seconds and it is usually right:

• Is this a skill I want to be world-class at in five years? → do it yourself, most of the time
• Is this a task I know I hate and will never want to do deeply? → automate, no guilt
• Is this a repeatable chore (emails, summaries, boilerplate)? → automate
• Is this a high-stakes piece of work where getting it wrong is expensive? → do it yourself, use AI as a reviewer at most
• Is this a domain where my taste needs to be sharp? → do it yourself for at least the first pass

The monthly AI-free audit

Once a month, pick a task in your core craft and do it without AI, start to finish. Time it. Notice where you struggle. Notice what you had forgotten. This is your audit.

The gaps you find are your real skill decay, not theoretical. Spend the next month closing the specific gap you identified. One gap per month. Over a year, that is twelve recovered skills you would have lost silently.

Rules for when I do use AI

• Never paste its answer without reading every word
• Never send AI-written text without changing at least three things — keeps my voice
• Always ask "what would I have written?" before I see the output. Forces thinking
• Keep a journal of times AI was wrong and I noticed. This protects my detection instinct
• Never use AI for the first iteration of a creative idea — the first idea is where my taste lives

Why most people fail at this

Not because they lack discipline. Because they never draw the line explicitly. Without the line, the default drift is always toward more AI — it is easier, faster, feels productive. In a year you are 95% automated, in skills you used to care about, and you did not notice the slide because every individual step felt reasonable.

The line has to be drawn on purpose and reviewed monthly or it moves without you. That is the whole game.

A rough working split

For reference, here is my current split across a workweek. Yours should differ based on your craft and goals.

• Writing long-form (essays, strategy docs): 100% me, AI only for copy-edit
• Code for my own products: 60% me, 40% AI-assisted (boilerplate, tests)
• Client code in domains I know well: 40% me, 60% AI-assisted
• Admin / routine email / scheduling: 10% me, 90% automated
• Research and synthesis: 50% me, 50% AI — I always read the primary sources
• Learning a new framework: 100% me for week 1, then 50/50 after

Skill is maintained the same way a plant is: small, boring, consistent watering.

The framework is simple. Following it is not. The reward for following it for five years is being a rare kind of professional in 2031 — one whose core skills are still sharp, who uses AI well, and who cannot be replaced by someone half as skilled with the same tools. That is the career bet.