Imagine your work software. The tabs, the fields to fill in, the buttons, the drop-down menu you have to hunt for three times before you find it.
Now take all that away.
That is what Salesforce presented this week at its annual conference: a version of its tools that works without an interface. The software is still there, it still does the same thing. It is simply no longer you doing the clicking.
How it works
Conventional software is built in two layers. Underneath, the logic: what calculates, records, checks. On top, the interface: the screens that let you drive that logic.
The interface exists only because a human needs to see and to click. A machine has no use for it. It can talk directly to the layer underneath.
Removing the interface therefore means acknowledging that the main user is no longer a person. It is an agent, which receives an instruction in plain English and triggers the corresponding operations.
In practice: instead of opening a customer record, changing three fields and saving, you write what you want. The agent deals with the three fields.
An interface is not just a convenience. It is what shows you the state of things. When you open a dashboard, you happen to see figures you were not looking for, an anomaly, a forgotten file. That accidental discovery disappears when you only ever look at the answers to your own questions. You get exactly what you ask for, and nothing else.
Why every vendor is going there
Salesforce is not alone, and the reason is simple: their customers are asking for it.
A company running agents wants them to be able to act inside its tools. If the agent has to simulate clicks in an interface designed for humans, it is slow, fragile, and it breaks with the slightest visual update.
Opening up direct access is cleaner. And it shifts the security question: it is no longer "who can log in" but "what is this agent allowed to do", a subject we covered in detail in our permissions guide.
The real change is elsewhere
We are used to judging software by its usability. Is it clear, is it fast, how many clicks does X take.
If nobody looks at the screen any more, those questions lose their meaning. What matters becomes: did the agent understand what I wanted, and did it do what was needed?
Those two questions are far harder to check. A badly placed button is visible. A misinterpreted instruction is not — it produces a plausible result, and you will only find out it was wrong by cross-checking it.
What it means for you
If you work with this kind of tool, three things will change, probably in this order.
You will write instead of clicking. Formulating becomes the core skill, and an imprecise instruction costs more than a misclick.
You will check more. Time saved on data entry will be partly taken back by verification. That is not bad news, provided you plan for it.
You will lose the glance. That diffuse knowledge you acquire by navigating a tool for years. It cannot be replaced, and nobody has yet proposed an equivalent.
This is not the first time a screen has been removed. Switchboard operators saw the whole telephone network pass before them. When connection became automatic, nobody had that overview any more. The service got better. The understanding did not.