Will AI Kill the GUI? No — and Here's Why!

• The case for yes
• But then there is the other app on my desk
• So no, AI will not kill the GUI
• What this means for building

I can do almost everything in a chat box now. I give an instruction, the agent does the work, and the result comes back. So the obvious prediction is that the graphical user interface is finished. No more screens, no more dashboards, just a conversation.

I do not think so. And here is why.

§The case for yes

Start with the app that makes the “yes” case look strongest. Mine.

deepDive takes two inputs from me: a Python payload with the research, and a base audio file. It gives me four outputs: a published article on the web using mdBook, a long-form podcast episode saved to my Google Drive, clips saved to GitHub so my distribution agent can pick them up and publish them, and a long-form video combining the clips, which the agent distributes to Spotify, YouTube, and TikTok.

As a user, I do not really care what happens in between. I do not care about locking or not locking, what the intermediate steps are, what the visual artifacts look like, or how good the UI looks. None of that is the real issue. The real meat of the application is two inputs in, four outputs out.

The real question is: are the outputs good? Exactly as I wanted? If not, I adjust the pipe code. I might still use the UI to see what went wrong, but again, as a scaffold, not as the product.

The interface of this app is a development artifact. I use it while I am working with the application to develop the pipeline. I want to see the prompts. I want to see how the application is working. Over a period of time, with multiple iterations and bug fixes, I finally understand as a human what the true pipeline is. I customize the inputs and the outputs in a way the pipeline supports. I make improvements in the code.

And ultimately, the UI vanishes. I would never use it again.

The future of this pipeline is a Docker implementation. I provide the inputs in a Google Drive folder, and the outputs are available back to the distribution agent, wherever they get saved, my Google Drive or GitHub. There is no screen in that picture at all.

If every app were like this one, the GUI would be dead already. The chat box would have killed it.

§But then there is the other app on my desk

My portfolio app is a very different beast.

It is a way to show my portfolio to me. It is the interface through which I understand my assets and how they are behaving against the market and against a fixed allocation strategy. I need to look at this interface on a daily or a weekly basis to really understand where I stand.

We are not taking anything from the portfolio app. We might take some action, say rebalancing on Schwab or Vanguard, but at the end of the day it is a way to understand what my net worth is. The graphical user interface is a way to understand the world.

In this case the user interface is absolutely required, because I use that interface to understand reality. It is not a pipeline. The UI is the real deliverable.

No chat box replaces that. I do not want a paragraph telling me my net worth. I want to see it: where I stand, against the market, against the plan, in one view.

§So no, AI will not kill the GUI

It will kill one kind of GUI and leave the other standing. That is the whole answer.

There are two kinds of apps. Pipeline apps, and understanding apps.

A pipeline app’s value is in its outputs. The screen in the middle is scaffolding. It exists so a human can watch the machine being built, and once the machine runs, the scaffolding comes down. AI is very good at taking scaffolding down, because the chat box is a better way to say “run it” than a dashboard ever was.

An understanding app’s value is in the view. There is no output beyond the seeing. The screen is the product. AI cannot kill that, because there is nothing to replace it with. A summary is not a view. A human understanding where they stand needs to see it.

Most teams have this backwards. They gold-plate the dashboard of a pipeline nobody will look at in a year, and they under-invest in the view a human actually lives in every day. They polish the scaffolding and neglect the mirror.

§What this means for building

If you are building a pipeline, do not polish the UI. Spend the effort on three things: the determinism of the pipe itself, the handoff points where inputs go in and outputs come out, and observability while you are developing it, so you can see what it is doing. That is what the interface is for, and only while you are building.

If you are building an understanding app, the UI is the product. Every pixel earns its keep, because the pixel is the point. A chart that misleads is not a cosmetic bug, it is the product lying to you about reality.

This is the same split I keep coming back to. Agents should be a thin decision layer over best-in-class deterministic software. The chat box is where decisions happen. The GUI survives only where a human needs to see reality, not where a machine needs to do work.

So the next time someone shows me a beautiful dashboard for a pipeline, I will ask one question: will I still be looking at this in a year? If the answer is no, it is scaffolding. Build it cheap, and spend the effort on the pipe.

The GUI is not dying. It is being sorted into the two things it always was: the scaffolding that comes down, and the mirror that stays.