Copying Code Into a Chat Box Is Not a Workflow

The annoying part now is not whether the model is useful.

It is.

The annoying part is the ritual around it.

Copy the code. Trim the code. Explain the missing context. Paste more. Realize you left out the one file that actually matters. Paste again. Get the answer. Drag the answer back into the real work. Repeat until your patience quietly leaves the room.

That is the workflow.

It works. That is not the same as saying it works well.

This is why the conversation has changed for me. A few months ago the interesting thing was that the model could help at all. Now the more interesting thing is that it is helpful enough for the wrapper to become the obvious problem.

That is a very different stage.

The limitation here is not intelligence. The limitation is the surface. The interface still lives outside the work, which means I am still hauling context back and forth like some kind of miserable courier for my own codebase.

That is not a serious long-term workflow. That is a workaround.

A useful workaround. Still a workaround.

And if the workaround is useful enough that I keep coming back to it, then the next meaningful leap is not just “better answers.” The next leap is whatever stops forcing me to manually ferry everything into a detached chat box just to have a conversation with the tool.

That is the real signal now.

The old search loop was the first obvious waste.

The chat box is starting to become the next one.

And once that becomes obvious, you stop caring only about whether the model is smart enough. You start caring about where the interaction lives, how much setup it takes, and how much dumb labor happens before the useful part begins.

Because at some point the answer is no longer the bottleneck.

The box is.