Scared AI will replace the hands-on work you’re good at? Ten minutes turns a vague fear into a real, personal list of what to learn next.
1. Pick one task you did by hand
Choose something you actually built this month without AI: a script, a campaign, an analysis, a post. Not a hypothetical, a real finished thing. Example: a script you wrote to dedupe customer records.
2. Write down exactly what you built
Paste the actual code, email, spreadsheet steps, or draft into the chat. Don’t summarise it, give the AI the real thing so the comparison is honest.
3. Ask AI for the architecture version
Say: ‘Here is how I solved this by hand. If you were designing this at a system level instead of hand-building it once, what would you design, and why.’ Let it answer in full, even if it feels abstract.
4. Make AI diff the two
Ask directly: ‘Compare my manual solution to your system design. List exactly what decisions you made that I didn’t, and name the skill behind each one.’ This is the step that does the real work.
5. Turn the diff into this week’s list
You’ll get 3 to 6 named gaps: things like idempotency, error handling at scale, or campaign sequencing. Pick two, not all of them, and find one video or one practice task for each.
6. Repeat weekly with a new task
Do this once a week with whatever you built that week. After a month you have a roadmap written from your own real work, not a generic ‘skills of the future’ list someone else made up.
A worked example
A software engineer manually wrote a script to find and remove duplicate customer records before a weekly report.
Before
A 40-line Python script: read CSV, group by email, keep the newest row, write output. Runs once a week when I remember to.
After
AI’s architecture: a small service triggered on new data, idempotent so it can rerun safely, with a dead-letter log for records it can’t confidently merge, and a dashboard metric for duplicate rate over time. The diff AI gave: ‘You solved correctness. I added idempotency, observability, and failure handling, those are system design skills, not scripting skills.’ That became the engineer’s three-item learning list for the month.
Three ways to bend it
Marketing manager
Compare one campaign email you wrote by hand to the ‘campaign system’ AI would design: triggers, segments, and a feedback loop instead of a single send. The diff shows you what ‘always-on’ marketing actually requires.
Junior data analyst
Compare a one-off spreadsheet analysis to the data pipeline AI would build for the same question. The gap usually points to automation and data validation, not formulas.
Freelance copywriter
Compare a single blog post to the content system AI would design across formats and channels. The diff often points to repurposing and distribution, not better sentences.
When it goes wrong
AI’s answer is full of jargon I don’t understand
Ask it to explain the architecture using only words from your current job. If it can’t, it hasn’t actually understood your work yet, give it more context and ask again.
The diff feels fake, like it just padded the answer
Pick a bigger or messier recent task. Simple tasks don’t have enough architecture hiding in them to produce a real gap.
I get a long list and don’t know where to start
Only ever take two items per week. A ten-item list is a reason to quit, not a roadmap.
When not to use this
If you haven’t done real hands-on work in this field yet, you have nothing honest to compare, learn the fundamentals first, then come back and run this on your own output.
What you can now do
You stop guessing what ‘AI-augmented architecture’ means for you and get a living roadmap built from diffs on your own real work.
Part of Making Life Easier With AI, a weekly walk-through of one small thing AI can take off your plate.
