Can LLMs replace software engineers?

No — LLMs can't replace software engineers wholesale, but they're already replacing a good chunk of what junior engineers used to do all day. The honest way to describe where things stand: the job is shifting from "write every line yourself" to "describe what you want, review what the AI produces, and integrate it correctly."

What are coding assistants actually good at?

Coding assistants genuinely earn their keep on a specific set of tasks. They're fast and reliable at boilerplate, translating code between languages, writing test suites, explaining a function you've never seen before, and turning a clear spec into a working first draft. Hand one a well-scoped ticket and it'll often finish it faster than a person would.

The split is fairly clean:

• Good at: boilerplate, translating between languages, writing tests, explaining unfamiliar code, first drafts from a clear spec
• Weak at: messy undocumented codebases, architectural tradeoffs, novel or subtle debugging, owning what actually ships

Where do LLMs still fall short?

The hard part is a real codebase that's five years old, half-documented, and full of decisions nobody wrote down anywhere. Understanding that mess, weighing tradeoffs against business needs, and taking responsibility for what goes to production is still a human job. LLMs can also produce code that looks correct and isn't, especially on subtle bugs nobody has seen before — which means someone has to read every line closely enough to catch it.

So is the job disappearing or just changing?

Changing. Junior and boilerplate-heavy roles feel the most pressure, because that's exactly the work assistants do well. Senior engineering judgment — knowing which tradeoff to make and why, and answering for the result — is more valuable, not less.

The center of the job moves from typing code to specifying it, reviewing it, and integrating it correctly, so the engineers who thrive are the ones who can direct the tool and still vouch for what ships.