Every help system does the same unhelpful thing: it tells you which button to press. If you could find the button you would not have asked. This one takes the wheel and does it on the page in front of you.
Five real systems, rebuilt faithfully — a hospital, a university, the railway booking flow, a shop, and a Japanese care home’s incident report. Pick one, try to do the job yourself, then hand it over: you hear what the agent is about to do, then you watch it happen on the screen in front of you. Nothing is submitted anywhere.
मेरिडियन प्रौद्योगिकी संस्थान
Meridian Institute of Technology
Before you press startTry to find the re-evaluation form yourself.
It drives the page while you talk. Nothing is submitted. Escape stops it.
Not a chat window that quotes the manual back at you. An agent with its hands on the page you are already looking at.
“I need to change the card on my account.” No menu names, no jargon, no guessing what the feature is called on this particular site.
Scrolling to the right place, opening the right thing, filling what needs filling — on the page in front of you, where you can watch it happen rather than being sent a link to somewhere else.
It narrates the route as it goes, so you learn where the thing lives instead of just arriving there. Next time you will not need to ask — which is the opposite of what a support ticket teaches you.
Not a search box guessing at keywords. It finds the form that lives under Examinations, three clicks past where anyone looks, because it reads the page the way you do.
Say wait, or press escape, and it stops. It is driving your screen, so nothing gets submitted without you watching it happen first.
It reads the page as it is — no restructuring, no tagging every button, no six-month integration project before anyone can try it.
Long applications lose people in the middle — a field they do not understand, a document they cannot find. The agent walks them through it in the moment, instead of you finding out later that they abandoned it.
Every mature product accumulates features its own users never discover. Asking for something by description rather than by location makes the depth of the menu tree stop mattering.
Written for a version of the interface that has since changed, and useless to anyone who cannot map the words to what is on their screen.
Some portals strip out every affordance in the name of being simple and end up with a wall of identical grey links. Asking for what you want routes around the simplification instead of fighting it.
Internal tools are where the training budget goes and the shortcuts multiply. Talking through a task beats remembering which of eleven screens it lives on, especially for the person who does it twice a year.
It can mention the thing that is selling out, or the size that actually fits, the way a good assistant would — and it asks before opening anything. Nothing is added to the cart because an algorithm decided you wanted it.
It drives a real page today, on the same engine as the calling agents and the 3D avatars that run across Singapore’s public sector. Point it at the journey your users struggle with and watch it work.
It drives a real page, talks while it works, and sees through the camera. It runs on the same engine as the calling and avatar agents — the part that is new is the hands, not the conversation.
Somewhere with a form people abandon, or a feature buried where nobody finds it. That is where the difference is obvious, and it is the kind of problem we would like to be measured on.
Tell us the journey people struggle with. We will put the agent on it and let you watch it drive, before anyone talks about price.
Name the journey. We will put an agent on it and let you watch it work.
The other two, and us