Why is it, that in 2025, we take it for granted that most web experiences are bad?
Booking a doctor’s appointment. Renewing your passport. Paying taxes. These interactions are typically frustrating, confusing, and error prone. For many, like the elderly, they may be completely impossible.
This isn’t mere inconvenience. Every organization with bad experiences is getting less of what they want, or paying a higher cost for what they want.
Whatever your mission is – from raising money for charity, to selling donuts – your organization is less effective with a bad UX. You may even fail because of it.
Right now we live in a world where only a slither of tech-savvy brands – like AirBnB, Apple, or Amazon – seem able to offer great user experiences. Almost everyone else doesn’t.
We believe the world can do better.
Problem 1: Users are screaming into the void
The vast majority of organizations have no awareness of what their users think. Most don’t even pause to consider them.
For those who do, surveys, interviews, and polls are blunt instruments. These are valuable tools, but slow and imprecise. And they almost always occur at the worst possible time: after a thing has been built.
Like with web accessibility, people know they should do, but they sacrifice it due to time, or budget, or lack of interest. And once a thing is built, who really cares?
Problem 2: Most tech solutions are actually tech problems
There’s a paradox in technology, where as we improve, we add more complexity. And that complexity can nullify the improvement.
For example: internet access gets faster every year. But we’ve bloated our webpages so much that speed remains roughly the same.
In the same way, we now have thousands of technologies purporting to make web development easier: React, Tailwind, Figma, CI, WebSockets, and TRPC.
Yet web development keeps getting harder. And there’s no end in sight.
Rethinking what is possible
We believe recent advances in AI can unlock dramatically improved user experiences.
Core to our plan, is what we call “synthetic user testing”.
Modern AI can already take control of computers like a person, using their ‘eyes’ to interpret the screen, interacting with a web browser, and pursuing a goal to completion. Multiple demos existed in late 2024.
We propose taking this further, to the point where website owners can deploy hundreds of ‘synthetic’ users to test any aspect of a website, each like a real person. These agents will each embody a distinct but internally consistent identity, with a synthetic psychology, history, biases, and input modalities.
AI could impersonate a teenager on their phone, or a senior with limited computer experience.
The aim is to dramatically decrease the time taken to obtain diverse, high-quality user feedback. Currently it takes hours to perform simple end-user testing; for a proper study it might take weeks or months.
We believe some close approximate of this process could occur virtually in real-time, during the design and development process. Swarms of simulated users could experience the equivalent of weeks of testing in minutes.
The people working on user experiences can obtain guidance from their intended audience, as they design and build.
Is that even possible?
It certainly isn’t easy.
Traditionally AI has been taught to behave as well as possible, which actually works against us. For our plan to work, we need AI with accurate human foibles: AI that can get bored, frustrated, confused, and give up.
Simulating this with enough accuracy is a hard problem, but we believe it is solvable. Co-ordination, consistency, and performance are all concerns.
But even a partial solution could offer a tremendous amount of benefit. And we think the tailwinds of modern AI will expand what is possible at hereto unimaginable rates.
Advice is not enough
Even if every organization in the world had access to near-instant user feedback, many wouldn’t have the resources to accommodate them. Change is expensive.
We anticipate AI will change this.
Generative models have already proven capable of coding, writing, designing, and even making video; albeit at inconsistent levels, and with human assistance.
It’s still early days, but it’s already clear that the future of all tooling is AI-assisted. Design tools, coding tools, writing tools – all of these will soon be able to understand and work towards specific goals.
But what they won’t do by default is have a robust understanding of our end users. This is where we intend to fill a gap: by integrating our in-depth model of your audience into the AI-powered tooling of the future.
Breaking out of the Stone Age
Even with AI assisted development, many of our most important organizations are hamstrung by ancient technology they can’t escape.
The US Internal Revenue Service, for instance, runs on a computer system developed in the 1960s, written in assembly language and COBOL, and its replacement isn’t anticipated until 2030.
One potentially exciting application of AI is layering better UX on top of these systems. Rewriting them is too slow, and connecting to them directly, via APIs, is currently a complex and brittle.
But AI could proxy between old-and-new interfaces, similar to a team of humans, but near instantly and without the cost.
In this way, we might be able to build high quality user experiences without waiting decades for the systems behind them to be ready.
The future that we want
We believe user experiences should be delightful by default. One day, even government websites, may become as pleasant and effective to use as AirBnB is today.
For this to happen, two things appear necessary:
- Near instantaneous user feedback, at all stages of development
- AI accelerated production
Today great user experiences are rare not because they’re impossible but because they’re inconvenient.
By making greatness easier, we believe it can one day become expected.
