Can task-oriented UI patterns reduce the friction of AI chat interfaces?

Friedman, writing for Smashing Magazine, argues that "chatbots are rarely a great experience paradigm — mostly because the burden of articulating intent efficiently lies on the user." He cites usability tests where "users often get lost in editing, reviewing, typing, and re-typing," taking "30-60 seconds for input." His proposed fix is not to remove chat but to study "5 key areas that we need to shape" when designing an AI feature: how users express intent, how output is presented, how users refine results, how agents take on delegated tasks, and where in a workflow the feature lives.

For intent, Friedman points to tools that sidestep typed prompts: Flora AI lets users "visualize their intent with nodes by connecting various sources visually," and Krea.ai lets users "abstract away the object we want to manipulate" and move it directly on a canvas instead of describing it in words. For output, he points to Amelia Wattenberger's PenPal, which adds "style lenses" (sentence length, sad–happy, concrete–abstract) instead of a plain text block, and to map-based outputs with togglable data layers. For refinement, he cites Adobe Firefly's "knobs, sliders, buttons" and contextual prompts that act on a highlighted selection "rather than global prompts." Agentic tasks let AI "initiate tasks... on their behalf," such as scheduling or research; workflow embedding places AI inside tools people already use daily — Slack, Teams, Jira, GitHub — rather than a dedicated AI tab, with browser autofill offered as the model to imitate instead of "a frightening and omnipresent text box."

The five-area framework runs parallel to How should AI interfaces handle the shift from doing to supervising?, which formalizes a similar intent-to-manipulation path into three named surfaces; Friedman's version is a practitioner's working checklist built from shipped product examples rather than a layered architecture. It also matches the direction of evidence in Do generated interfaces outperform text-based chat for most tasks?, where task-specific generated UI beat chat in a controlled comparison — Friedman's intent- and output-shaping areas describe the design moves that produce exactly that kind of task-specific interface. And it shares Is the AI capability gap really an interface problem?'s diagnosis that the bottleneck sits in interface design, though Friedman supplies concrete patterns (canvases, lenses, highlighted-selection prompts) where Mollick argues the point in the abstract.

Friedman's piece is a design essay built on his own observation of shipped products and an unspecified set of "usability tests," not a controlled study — the 30-60 second input-time figure and the claim that users "often get lost" are not sourced to a named study or sample. The five areas are offered as a checklist for practice, not a tested taxonomy with validated boundaries between categories. The implication, at the strength the essay supports, is that teams designing an AI feature gain a concrete starting checklist, not a validated causal account of why task-oriented UI outperforms chat.