Smashing Magazine: Users don't need more tools, they need quiet integrations

Vitaly Friedman, Smashing Magazine’s editorial director and a longtime UX practitioner, uses this piece to push back on a pattern that has dominated AI product development: the assumption that users want a new application to add to their stack. His argument is simpler — people already have tools they understand and workflows they’ve built around those tools. Adding another interface creates friction. The opportunity is in reducing friction by integrating into what already exists.

The concept he develops is “Quiet AI.” Where most AI products demand attention — prompts to write, interfaces to learn, outputs to review — quiet AI handles tasks in the background without requiring the user to shift mental context. His clearest example is Claude’s integration in Microsoft Word and Excel, which offers writing suggestions and data analysis without pulling users into a separate application. The value appears where the user already is.

Friedman also introduces a pattern he calls folder instructions, where a user defines an intent once and the system carries it out automatically going forward. He describes several use cases: a folder where attached documents get summarized by category, an invoice folder that routes files to the right client bucket, a renewal workflow that collects the right documents when a passport needs updating. These are high-frequency, tedious tasks that users would handle manually in the absence of any AI integration. Folder instructions turn a one-time configuration into ongoing time recovery.

The broader design principle at work here is friction reduction over feature addition. Most tools grow by adding capabilities, which means growing the cognitive surface users must manage. Friedman argues that the most valuable AI design question isn’t “what could we add?” but “what does the user currently dislike doing, and can we make it happen without them noticing?”

The article is most useful for designers and product managers building AI features into existing platforms, or evaluating whether a standalone AI product is actually justified. The test he implicitly proposes: if a user would have to open a new application to get the value, ask whether embedding it in the tool they already use would produce a better outcome.