I Stopped Asking AI to Write Files and Started Giving It Tasks

When I first started coding with ChatGPT, the workflow was very literal. I would paste a file, explain the change I wanted, get some code back and then manually copy it into the project.

That already saved a lot of time, but it also meant I was doing all of the coordination myself.

More recently I have started using coding agents that can work directly with a repository, and that changes the useful unit of interaction.

Instead of saying:

Rewrite this component.

I can say something closer to:

Add this feature. Inspect the existing implementation first, preserve the current behaviour, update the necessary files and run the relevant tests.

That sounds like a small wording difference, but it changes the entire workflow.

The agent can inspect related files instead of relying only on whatever I remembered to paste into the prompt. It can see how the project is structured, search for existing patterns and make coordinated changes across several files.

My job shifts from transporting code between windows to defining the task well.

That does not mean I give it an enormous vague instruction like “finish my app”. The best tasks are still bounded. I explain what outcome I want, what constraints matter and what should not change. Then I let the agent investigate how the repository currently works.

I still review the result. I still care about architecture, and I still correct it when it takes the wrong approach. The difference is that I no longer need to micromanage every edit.

This feels like the biggest change in AI-assisted development since I started using ChatGPT for coding.

The useful question is becoming less:

What code should I paste here?

and more:

What is the engineering task I want completed?

That is a much better level to work at.