Back to the future (of user interfaces) and why AI Agents won’t solve enterprise software’s problems

and maybe a teaser of what we have been building at Beyond Work

What is a user interface? It’s a way to translate our intent into action. The first tools were as simple as stones used to smash nuts so we could get the nutrients inside, but developed into sticks and axes and so on. The better the tool, the better its design to make us more efficient and translate intent into action.

With computers, the stone-to-stick evolution came from Douglas Engelbart and the mouse, heralding the graphical user interface so famously borrowed by both Steve Jobs and Bill Gates. Bytes became commands, commands became clickable pixels, and the personal computer revolution was born.

Terminal interfaces like DOS or Linux are complicated to remember for the regular user but it’s very easy to build a lot of functionality fast. Graphical user interfaces, or as we might choose to call them in the future, “artisanal user interfaces,” are all painstakingly handcrafted by designers.

Today, modern user interfaces are more commonly made by large libraries of UX components, and this is the reason that every single application today looks more or less similar. This is also the reason they are static. Changes require software developers and designers to create new code or handle convoluted configuration of fields and buttons.

In that context, it starts to make sense how little actual innovation and development there has been in enterprise software for the last four decades. With the immense cost of building, maintaining, and testing user interfaces; the more user interface, the costlier an application.?

By some estimates, roughly 60% of the code in a modern SaaS application is front-end or user interface code (including libraries), and an average enterprise application is often more than a million lines of code. On top of that comes maintenance costs as a multiple in $.?

It’s time to start thinking about what all that user interface is costing us. It’s a billion-dollar problem.?

Innovative tools like Figma have made it cheaper than ever to create concepts and design, but that’s only compounding the problem of ever more UI to build, maintain, and test. That vast economic scale of user interface design can also be seen in Figma’s acquisition by Adobe ( the original artisanal user interface maker) – the largest software deal of the last 12 months at a staggering $20bn.

UI is costly and making it cheaper is big money.

One approach that has gotten much excitement lately is agents; AI-driven software clicking buttons for you and entering data into applications. Take Adept, who raised $350m, or induced.ai, backed by Sam Altman, who are using LLMs to do just that.?

But agents are nothing new. Legacy RPA automation platforms like UiPath or BluePrism are essentially just that: Automated agents to push buttons on artisanal graphical user interfaces. So-called automation platforms then require more graphical user interfaces to manage how and in what order to push the buttons (an Inception of user interfaces, if you will).?

Agents don’t solve the basic economic problem of the cost of user interfaces, as we still have to make and maintain them – plus they bring new problems of their own. Now, we are just making them for robots to push instead. It is as silly as imagining a robot inside your Tesla, holding the steering wheel for its self-driving features. Is Johnny Cab from Total Recall the future?

All these approaches fail at the primary tool test from when we used stones to smash nuts. Have we not overcomplicated the problem? No one created trebuchets to smash nuts when a stone could do the job; it’s uneconomical.

At Beyond Work, we have a different vision. It may be more radical but it restores first principles of what we believe a user interface should be: a tool to solve a problem. User interfaces should be shaped around the problem, for the user, visible when you need them and not adding clutter when you don’t. Let’s call it generative UI, or as we have dubbed it, in Beyond Work PromptPlus.?

Imagine if you didn’t have to remember what page a particular button was on or how it worked, let alone train for three months to get your certification to add a field to a Salesforce form. ?

Imagine you could just tell your computer what you needed to do, and it generated the UX on the fly based on your preferences – and then it went away when it had solved your problem.

so time for the promised teaser and an example of how software might work in the future...