Here's what's happening when you accept an AI suggestion:
You get the code. You don't get the architecture.
Not because AI can't generate architecture—it can. But because architecture isn't a deliverable. It's a lens. A way of seeing the problem at the right level. And you can only develop that lens by working at that level.
When AI generates your code, you're operating at the code level. Your brain consolidates at the code level. Over time, the architecture level—the level that actually makes you valuable—quietly atrophies.
What AI Actually Takes from Senior Engineers
A senior engineer's most valuable asset isn't knowing how to write code. It's knowing what level to think at.
- System design—choosing the right abstractions
- API contract thinking—what should this interface hide?
- Scaling intuition—where will this break at 10x?
- Technical debt awareness—which shortcuts are worth taking?
These are abstractions. They're built over thousands of hours of looking at systems, making decisions, and living with the consequences.
AI generates code at the implementation level. It doesn't build your abstraction intuition. You build it by working at that level.
And when you stop working at that level—when every implementation detail is handled by AI—the abstraction intuition stops building too.
The Abstraction Ladder
This is what it looks like when fresh:
Senior engineers live at Levels 4–6. Junior engineers are building their way up from Level 1.
AI is compressing this ladder. Engineers are accepting code at Levels 1–2 without developing Levels 3–6. The result: a generation of engineers who can ship fast but can't decompose systems.
Three Symptoms of Abstraction Erosion
You can implement but not design.
You can take requirements and produce code. You can't take a vague problem and propose the right system structure.
The framework is the architecture.
When your primary tool is AI, the framework (Next.js, Django, Rails) becomes the architecture. You design within the framework rather than choosing the right framework for the problem.
You know it works, you don't know why it works.
AI suggestions work because they're trained on patterns that work. You can verify correctness but not explain why this approach is better than alternatives.
How Senior Engineers Are Protecting Their Abstraction Ladder
The engineers most concerned about AI aren't worried about their ability to ship. They're worried about their ability to think architecturally while shipping.
1. Design before, generate after.
Before any AI session, write the design doc or sketch in plain language. Explain the decomposition, the interfaces, the data flows. Then use AI to implement. The act of writing the design forces you to operate at the architecture level.
2. Implement one layer yourself.
Pick the highest-value layer of any system—the architectural decision, the core algorithm, the contract definition—and implement it without AI. Even if AI could do it faster. This layer is where your skills live.
3. Review in abstraction.
When reviewing AI-generated code, don't just evaluate correctness. Evaluate whether the approach is right. Does this belong at this layer? Is the interface appropriate? Is the decomposition clean? Make these architectural judgments explicit.
The Test
Can you describe the architecture of the last system you shipped—not what it does, but how it's structured—at an abstract level that someone without your codebase could understand?
If yes, your abstraction ladder is intact.
If it's hard to describe at that level, that's the signal. AI has been doing the abstracting for you.
A Note on Recovery
The good news: abstraction intuition doesn't leave. It goes dormant.
The practice is the same as skill retrieval. Work at the right level, and the fluency comes back. The architecture lens reopens.
It takes effort. It's slower than accepting AI suggestions. It requires tolerating the friction you hired AI to remove.
But it's also the thing that makes you the engineer people want on their team when the problem is genuinely hard.