Last week I told you readiness is a date, not a feeling. This week, the thing that happened when I stopped waiting and let AI do more of my actual work.

It did not take my job. It did something more uncomfortable.

I have been using agentic coding tools across my own projects for most of this year. Refactors, infrastructure changes, the boring glue code I used to put off. The output was good. Faster than me, most of the time.

Then I noticed a pattern in where it went wrong.

Every time an AI change caused a problem, it was in a part of the system I could not have explained properly myself. An IAM policy I had copied and never really understood. A caching layer I set up from a tutorial two years ago. The tool did not introduce the gap. It walked straight into a gap that was already there, and this time I could not blame the tool, because I had approved the change.

That is what I mean by exposed. AI is a mirror for how well you understand your own system. Where you are strong, it makes you faster. Where you are faking it, it makes the fake visible, usually in production.

Most advice about AI tools is about getting better output. Better prompts, better context, better models.

The advice nobody gives career changers: the bottleneck moves from writing the code to being able to stand behind it. Once the code is nearly free, the person who can explain why the change is safe is worth more than the person who can generate it. Interviewers have worked this out. "Walk me through why you did it this way" is now the whole interview.

So the question is not "should I use AI?" You should. The question is whether you can defend what it hands you.

Before I accept any AI-generated change now, I make myself say three sentences out loud. No notes.

What does this change do? One sentence, plain English, no jargon.

What could it break? Name one specific thing. If the answer is "nothing", I have not looked hard enough.

How would I know? The log line, the test, the metric that would tell me within a day.

If I cannot get through all three, I do not ship it. I go and learn the thing first, and then the change is mine. It takes two minutes and it has caught more problems than any linter I have run this year.

Do this for a month and your understanding of your own stack rises faster than any course, because the tool keeps taking you to the exact places you were avoiding.

I put the three questions on a single sheet with a one-month log underneath. Run every AI change through it for four weeks and the "gap found" column becomes your personal curriculum. Stick it next to your monitor. Direct link, no signup:

📥 Download the Three-Sentence Test (1 page PDF)

You don't have to scroll every AI thread, track every new tool, or watch every demo. 

The Rundown AI breaks it all down for you — the latest AI news, tools, and tutorials in one free 5-minute email every morning. 

Trusted by 2M+ professionals at Apple, Google, and NASA.

Sign up to start learning.

💬 Where did AI expose a gap for you? A service you use but could not explain, a change you approved without really reading it?

Reply with EXPOSED and the thing. I will point you at the fastest way to close that specific gap. I read every reply.

Know someone shipping AI code they cannot explain yet? Share your referral link. 1 friend subscribes, you get a free LinkedIn Profile Optimisation Checklist.

{{rp_refer_url}}

Shola

P.S. Next Tuesday: I hold 10 cloud certifications. If I started again I would sit 5 before the offer. Which five, in what order, and why the other five only made sense once I was already inside. 6PM UK.