I Let AI Write Most of My Code. Here's What I Still Had to Do.

hoang — Sat, 03 Oct 2026 18:23:32 +0000

AI can write code surprisingly fast.

It can generate functions, create API endpoints, write tests, explain unfamiliar code, and even help fix bugs.

So I started asking myself:

If AI can write most of the code, what is the developer actually doing?

After using AI more seriously in my development workflow, I think the answer is more interesting than "developers will be replaced."

AI is very good at writing code

For many everyday tasks, AI is faster than I am.

For example:

Instead of spending 30 minutes writing repetitive code, I can often get a useful first version in a few minutes.

That's a huge productivity improvement.

But there is a problem.

AI can give you code that looks correct.

That's not the same as code that actually belongs in your application.

The developer still needs to answer questions like:

These questions are much harder than generating a function.

One thing I have noticed is that AI makes debugging more interesting.

When AI generates 200 lines of code in seconds, you can also get 200 lines of bugs in seconds.

The ability to understand why something doesn't work becomes extremely valuable.

You still need to read the stack trace.

You still need to understand the data flow.

You still need to inspect the database.

You still need to understand the system.

And sometimes you need to tell the AI:

"No. Your approach is wrong. Let's rethink the architecture."

That's still engineering.

I don't think AI eliminates software engineering.

I think it changes where our time goes.

Before:


text
Idea
  ↓
Write code
  ↓
Debug
  ↓
Test
  ↓
Deploy