Invisible Design, Revisited.
PH—017
The best interface is the one the user never thinks about. A year of install work has sharpened that claim. Here are the four things we now delete by default, and why the best AI products refuse to show off.
I The Original Claim, Revisited
The original Invisible Design post argued that the best interface is the one the user never notices. The argument was correct and too general. A year of install work has narrowed it to four specific moves, each of which produces a measurable improvement in how installed systems are adopted.
Invisibility is not minimalism. A minimalist interface can be hostile. Invisibility is the property of getting out of the way of the user’s task. It is additive in some places and subtractive in most.
II What We Delete by Default
1. Confidence percentages on AI outputs.
The user is not a calibration tool. A confidence of seventy-four percent is less useful than a binary “I’m sure / ask a human.” We keep the number internally. We do not surface it.
2. “AI generated” badges on outputs that are entirely AI generated.
If the whole product is AI, the badge is noise. We badge only mixed-origin content (human reviewed, edited, partially machine).
3. Typing indicators that fake human latency.
Users recognize the fake. It reduces trust. If the system is ready immediately, show the answer immediately.
4. Chat interfaces where a form would do.
Chat is a powerful interface for open-ended tasks. It is a worse interface than a form for closed-ended ones. The best AI products know when to surface one versus the other.
III Where Invisibility Goes Wrong
Invisibility has a failure mode: systems that hide so well the user does not know they exist. This is an error. On regulated surfaces especially, the user needs to know an AI participated. On inspectable systems, the user needs to be able to see the inspection.
The rule we have settled on: invisibility for the mechanics, never for the responsibility. Hide the model. Surface the source. Hide the chain. Surface the decision. Hide the tool use. Surface the audit trail.
IV The Cost of Visible AI
Visible AI, the kind that reminds you constantly that it is AI, has a cognitive cost we have tracked in simulation. Users re-read more. They edit more. They escalate more. The product feels slower, even when the underlying latency is identical.
There is also a brand cost. Brands that foreground their AI signal newness. Brands that background it signal maturity. For most of the verticals we install in, maturity is the positioning buyers want. Wellness wants it. Luxury demands it. Public innovation requires it. Only consumer-creative can afford to foreground AI, and even there the novelty is a depreciating asset.
The move is to make the AI invisible and the quality impossible to ignore.