Goodbye, Menu Interfaces
I won't have to learn your software; your software will learn what I want; Reminder: Substack live at 11 AM today
[Reminder: I will talk at 11 AM today New York time with James Cham on Substack Live. James is with the venture arm of Bloomberg, and he just came back from the AI summit in Paris. That plus plenty more to talk about in the latest developments in AI. I believe that Substack Live is only available on the phone app. I will welcome all subscribers, not just paid. This time, anyway.]
Above is a screenshot taken from my accessing the Wall Street Journal. It is all menus. there is a row of choices at the top (“Latest, World, Business,…”), a five-column menu in the middle (this is actually way down near the bottom of the web page), and another set of menus at the bottom (“Dow Jones Products, “Barron’s”, etc.)
For all of these menus, I still do not see two of the functions that I have searched for in the past: submit a letter to the editor; and suspend print subscription (for when we leave town)1
With the advent of large language models, I expect to see menus die. As you know I think of LLMs as the next evolution of the human/computer interface.
In the near future, I will not have to figure out the WSJ’s menu. Instead, I will simply type in or say “I want to submit a letter to the editor,” or “I am going out of town next week, so don’t deliver the paper then.”
The LLM interface will have two advantages. A relatively minor advantage is that it saves users the trouble of learning menus.
The major advantage is that the user will not be limited by the “features” as specified by menus. Users will be able to make up features as they go.
From the WSJ, I might ask for a comparison of stories from the first week of Trump 1.0 to the first week of Trump 2.0. Or I might ask for a list of predictions made by various columnists over the past five years and how those predictions turned out.
I do not picture the WSJ providing me with a large language model. I think that I will use something generic, like ChatGPT or Claude today. The WSJ will get out of the business of software design altogether, other than setting up a database.
What the WSJ will provide is access to a database of its current edition and its archives. No doubt I will pay for this access.
This will be the pattern for all of the services now offered on the Internet. I will not access them by going to Web sites and pawing through menus. Instead, I will use ordinary language, processed by an AI tool, to obtain the services that I want.
Web sites will wither away. Web design will cease to be an occupation. The user interface will be a generic AI tool, not something that service providers have to think about.
Service providers will expose the data and the capabilities that they have. Consumers will use their AI tools to obtain what they want. They might pay the service providers a la carte or through subscriptions.
To take another example, suppose that you operate a service to deliver food from various restaurants. You maintain a database of restaurants, prices, food items, ingredients, descriptions, and photos. As a consumer, when I want food delivered, I give, say, Claude an idea of what I am looking for. Claude accesses your database, and Claude and I have a conversation that results in making an order.
As a consumer, I might come up with my own use for your food delivery service. “I want to host at my house a competition among the five best pizza restaurants in the area.” Claude works with your delivery service to make this happen.
Actually, your delivery service might be disintermediated. Every restaurant could expose its offerings, and Claude could search those, with no need for a central third-party database. And Claude could arrange for delivery, without using you. But I imagine that restaurant reviews would still come from a third party.
The point is that with AI, there is no activity that consists of “visiting the web page(s) of the restaurant(s).” My conversation with Claude does not require that.
The computer interface that we have all gotten used to, involving web sites and menus, is obsolete. With large language models, we can access information and services more easily. And as consumers we can come up with our own functionality, going beyond the imaginations of the service providers.2
Yes, I am an old fogey who still gets a dead-tree newspaper delivered. I remember from experience that to suspend delivery I click on “customer center” in the second column. I forget where you go to submit a letter to the editor.
I am reminded of The Venturesome Economy, Amar Bhide’s book about consumers as innovators.
Interesting article. As I was reading it, I was thinking that some companies deliberately obscure functions behind complicated menus so you don’t do them, such as cancel. The other I was thinking about was advertisements, where would they go? So many UI’s today are designed to maximize advertisement display. Given the prevalence of advertisement supported services. I thought this was an interesting question. Although they could probably just put them on the generic search page you describe.
The WIMP (windows, icon, menus, pointer) metaphor has worked for a long time and I suspect we won’t be saying goodbye anytime soon. The wimp metaphor is a discovery mechanism. You can’t discover purely through language if you don’t know what to ask.