Why Invisible AI is Up Next
Opinion
In Introducing Deep Softworks, I made the argument against the chat bot UI paradigm for AI agents. Here, I'll make my case for why AI should be invisible in consumer software.
Intelligence is abundant but accessing it is hard.
Mankind is living through an era of abundant intelligence. Today's open and closed-source AI models can reason, transform, and synthesize at superhuman breadth.
The bottleneck now, I believe, is the interface through which that intelligence is accessed.
A popular idea has emerged among leaders in the AI space that frames humans as “managers” of AI agents to whom tasks should be delegated.
I view this as an attempt to retrofit old labor structures onto our new reality and a fundamental misunderstanding of the medium.
Treating an LLM as a "person" to be managed - asked to do things - is anthropomorphism at its most inefficient.
When we treat a tool like a person, we inadvertently inherit the social baggage of human interaction: the labor of explaining context, and the expectation of a back-and-forth.
In design, this is a regression. We don't "negotiate" with a hammer to hit a nail, nor do we "delegate" a calculation to a spreadsheet. We simply use them.
By personifying AI, we keep it at arm's length, forcing a dialogue where there should be a direct execution of will. We can, and must, remove this "asking" step.
A seasoned driver spends none of their cognitive bandwidth thinking about the mechanics of the accelerator or the rotation of the steering column.
Instead, they are concerned with the flow of traffic, the curvature of the road, and the navigation of the route.
To the driver, the car is an extension of their own nervous system. A sort of "ready-to-hand" tool so perfectly integrated that the machinery disappears, leaving only the experience of movement.
The best tools are engineered this way. To recede into the subconscious. They become extensions of agency, empowering humans to do more, reach farther, and prioritize better.
AI, in its current consumer form, foregrounds itself. It asks to be addressed and at times celebrated with flashy buttons and annoying popups.
But intelligence should be ambient. Ever available at the exact locus of intent, as humans have always had it.
If AI is to become the infrastructure for thought, it must stop acting like a guest in our workspace and start acting like the floor beneath our feet.
For our approach to this design direction, see Rawa.