Lately I keep circling back to the same handful of questions. What AI actually means for junior engineers. How to use it well without hollowing out your own skills in the process. What is still worth learning when a model can produce working code in seconds.
So this week: no story, just a straight FAQ. Five questions, five honest answers.
1. Is AI going to make junior engineers obsolete?
No. What it does is compress the timeline, and it does that fast. The floor work, wiring up boilerplate, writing the obvious version of a function, is exactly what AI is best at right now. What is left standing is the part that was always the hard part: deciding what to build, judging whether a design will hold once real traffic hits it, seeing the failure mode before it ships. That work does not disappear. It just moves onto you sooner than it used to.
2. How do I actually use AI well, day to day?
Decide first, generate second. Work out the architecture yourself: what needs to be a service, what needs a queue behind it, where the data actually lives. Only then hand the implementation to AI, and read every line it gives you back the way you would review a teammate's pull request. Treat it as a fast, confident collaborator who is sometimes wrong in ways that look right. You are not outsourcing the design. You are outsourcing the typing.
3. If AI writes the code, how do I still build real skills?
By taking the reps AI cannot take for you. Watching a model produce a clean solution teaches you nothing, the same way watching someone else debug a production incident teaches you nothing. So design the system before you prompt for it. Write down where you expect it to break. Then compare your prediction to what the code actually does under load. Judgment gets built by deciding and sometimes being wrong, not by reading a tidy output.
4. My team expects more output now that everyone has AI. How do I keep up without burning out?
Stop competing on output. Output is the one thing that just got cheap. Let AI take the implementation grind, the boilerplate, the first draft, and reinvest the time you get back into the work that actually compounds: the design conversations, the tradeoffs, the code review. The engineers burning out right now are racing AI at the one task it is already better at. The ones who thrive are moving up to the scarce thing, which is judgment.
5. What should I actually learn to stay valuable?
Systems thinking. Underneath every application you use, there is a small, repeating set of building-block patterns: services that answer requests, queues that let you say yes now and do the hard work later, databases and stores that hold the data, workers that grind through it in the background. It's the same handful of patterns you can see if you look at how Instagram, Stripe, or Netflix were designed. AI changes the syntax it takes to write them every few months. It does not change the patterns. Learn those, and you can design a real system in any language, with whatever tool happens to be sitting in front of you.
The thread running under all five: AI brings the implementation. You bring the judgment. Learn the patterns, and the rest really is just syntax.
- Kay
P.S. If you know a junior engineer chewing on any of these, forward this their way. And if this landed in your inbox because someone forwarded it, you can get the Saturday letter yourself at systemthinkinglab.ai.
This letter goes out by email every Saturday. Subscribe here or read the full archive at systemthinkinglab.ai/newsletters.
Top comments (0)