Writing Code by Hand Is Over: AI Slop, Vibe Coding and the Real Bottleneck

Writing code by hand is over. It's never coming back. Including reviewing code. AI does it better, faster and cheaper than us.

I know that sentence makes a lot of engineers angry. I'm an engineer too. But I'd rather say it plainly than pretend the last few years didn't happen.

Writing and reviewing code by hand is over

For most of the history of our industry, the expensive part of software was typing it. You needed people who knew the syntax, the libraries, the frameworks, the edge cases, and you needed a lot of their time. Then you needed other people to read all of it again and approve it.

That is no longer the expensive part. AI writes the code, and AI is also better at reviewing it than a tired human going through a huge diff at the end of the day. Clinging to manual writing and manual review is not quality control. It is habit.

So focus on what universally matters more: creating value, solving problems, being useful.

AI slop is a synonym for incompetence

AI shaming or just resisting change?

When people say "AI slop", they usually mean bad output that was produced with AI. Fine. But bad output was around long before AI. The tool did not create the incompetence, it just made it faster and more visible.

AI slop is just a synonym for incompetence, not AI being incapable. If someone ships something sloppy with an AI tool, the problem is the person who did not know what good looks like, did not check, and did not care. Blaming the model for that is like blaming the compiler for a bad architecture.

Vibe coding versus the average developer

Vibe coding gets the same treatment. Every week there's a post showing some terrible piece of AI generated code, and the conclusion is always that vibe coding is dangerous.

My answer: vibe coding, on average, produces better results than if average people were writing code manually.

That's the comparison that matters. Not the best AI output against the best senior engineer on their best day, but average against average. And the average hand written codebase was never the masterpiece people now pretend it was.

The real bottleneck: legacy mindsets and processes

I recently replied to a post arguing that AI is getting very good at writing code, but not at shipping it. I agree with the observation, not with the conclusion.

The problem is legacy mindsets and processes that are the bottleneck, not AI.

Most companies still run on a process designed for a world where code was slow and expensive to produce: tickets, handoffs, review queues, approval chains, release trains. If you plug a tool that writes code in minutes into a process built around code taking weeks, of course nothing ships faster. You just moved the waiting somewhere else.

That's not a limitation of AI. That's an organization that hasn't changed yet.

Focus on value, problems and being useful

None of this means engineers are useless. It means the value moved. Nobody pays for lines of code. They pay for a problem that goes away, a product that works, a process that's faster.

So stop defending the craft of typing. Learn what good looks like, use the tools properly, and fix the process around them. That's where the work is now.