“Developers will stop opening code editors altogether.”
I said that years ago.
At the time, it sounded like a bold prediction. Lately, I find myself thinking about it often. Something close to that future has started to emerge, although it looks a little different from what I imagined.
Work now starts with the right Option key
I recently configured Codex on macOS so that pressing the right Option key starts voice input. No matter which app is in front of me, I can press one key and start talking.
When something goes wrong, the first thing I open is no longer my IDE. I press Option and say:
“Find the root cause.”
“Check the official documentation and fix it.”
“Run the tests too.”
Codex searches the codebase, checks the documentation, edits files, and runs commands. I review the changes, the test results, and the actual interface, then decide what should happen next.
The editor still has not opened.
At first, I thought voice input had simply become more convenient. But if voice only turned my speech into text, it would not have changed the way I work. Now, my words do not stop at transcription. They continue into file changes and test runs.
Even when I do open an editor, it happens much later. Some tasks are completed without opening one at all. A scene I once imagined is gradually becoming part of my everyday work.
What decreased was not typing, but translation
To make a computer do something, I used to translate what I wanted into search queries, file paths, commands, and code. Even when I knew the desired outcome, I still had to convert it into a procedure the computer could execute.
AI agents take on part of that translation. Instead of specifying every operation in sequence, I start by explaining what seems wrong, what outcome I want, and how far the investigation should go.
The idea that a person sets the goal while a system handles part of the execution is not new. Back in 1999, research on mixed-initiative interfaces argued that direct human control and agent-driven automation should work together. What was once discussed mainly as a research direction is now becoming tangible in everyday development tools.
That does not mean input disappears. The more work I delegate by voice, the more output I need to inspect. A user study of generative coding tools similarly found that AI could help people get started, while understanding, editing, and debugging the generated code remained separate work.
The bottleneck has not vanished. The cost of specifying procedures has gone down, while the work of explaining intent and validating results has moved to the foreground.
Delegating execution is not the same as delegating judgment.
Thinking beyond voice
When I think about what might follow this shift, I think of brain-computer interfaces, or BCIs. Voice still requires us to turn thought into language and then transmit that language as sound.
That does not mean BCI will simply become the next input method after voice. One representative brain-to-speech interface study decoded signals produced when a participant with ALS, who had implanted microelectrode arrays, attempted to speak. That is still far from technology that can freely read a person’s thoughts.
I have no idea when the next major transition will happen. But I suspect we will keep looking for ways to communicate intent to computers through fewer layers of translation.
Revisiting the prediction
I still open my editor. Sometimes I need to follow the code directly in order to think. For a small change, editing it myself can still be faster.
What has changed is when I open it.
Most of my work used to begin in the editor. Now I start by telling an AI what I want. Once it produces a result, I enter the editor only if I need to.
The era of not opening a code editor may not arrive because editors disappear. It may arrive because more and more tasks can be completed without opening one.
Every time I press the right Option key, I am reminded that the future I once described may not be as far away as it sounded.
This is an English adaptation of my original Korean essay. I used AI to help with translation and editorial refinement; the ideas, examples, and final responsibility are mine.
Top comments (0)