Op-ed: vibe coding is fine, actually
the panic about AI-generated code is the panic about every new tool. we have been here before. here is what i think actually changes.
every few years, the dev community decides that a new tool is going to destroy software engineering. the panic is always the same: "we will not need engineers anymore."
i have been through this. i was a junior when people said Ruby on Rails would end CS degrees. i was mid-career when people said jQuery would end frontend engineering. i was senior when people said Kubernetes would end sysadmin. none of those things happened, and none of the engineers who actually knew how to use the tools lost their jobs.
what is actually different about LLMs
the new thing is not "AI can write code." we have had code-generation tools for decades. the new thing is that LLMs can write code in your codebase, with your conventions, at the speed of conversation.
this is a real capability change. it is not a category change. engineers will still be needed; we will just be doing more of the "what should this code do" work and less of the "type the code" work. that is a better job.
what i think actually breaks
- the bottom of the market: interns and FYP candidates doing CRUD work. if you are paid to write 200 lines of JavaScript a day, an LLM will do that. the answer is to learn the parts the LLM cannot do (architecture, design judgment, asking the right question), not to "code harder."
- the "interview is a leetcode" model: if your hiring process is "can you recite the answer to FizzBuzz," it is now obsolete. good. switch to project-based hiring.
- the "i will just write it myself" ego: the engineers who refuse to use LLMs will be the ones who ship less, not the ones with purer souls. use the tools.
what does not break
- the need for someone who understands the domain.
- the need for someone who can debug a system that is on fire.
- the need for someone who can hold a 6-month roadmap in their head.
if you are an engineer reading this and you are worried about your job: stop writing code, start writing intent. the LLM can turn intent into code. it cannot yet come up with the right intent.