Interaction with assistants that execute actions generates an expectation that goes beyond chatbots: why can’t a traditional application execute the requested operation, instead of showing a path to follow on your own? The user describes the desired result (“not a list of flights, but an itinerary compatible with my constraints”; “not the list of documents for an application, but the completed application”) and delegates the orchestration: they define the output, they do not manipulate elements one by one.
Four consequences, with very different levels of maturity — the first two are design changes already visible, the third is a constraint to design, the fourth is an effect documented by research:
- Already observable. Intent-based CTAs replace function-based ones: “Pay a bill” instead of “Payments”. It is not a label change but a perspective change: from what the system offers to what the user wants.
- Observable in AI platforms, still rare elsewhere. The search bar no longer only serves to find, but to command: “Approve Mario’s expense report”, “Create a project for client X”. At I/O 2026 Google redesigned the bar in this direction — it expands while typing, accepts text, images, files, and video, anticipates intent — calling it the biggest update to the search box in over 25 years. In traditional products the pattern is still experimental: it is a direction to evaluate, not a requirement.
- Constraint to design, not automatic consequence. Delegation and trust grow, but not uniformly: gladly on informational or low-criticality tasks, high threshold on payments, medical decisions, legal acts. It is risk calibration, not irrationality; but the threshold varies greatly between people and generations, and designing it without paternalism or naivety is difficult.
- Effect documented in research. A cognitive risk grows, now measured: a Microsoft Research and Carnegie Mellon study of 319 knowledge workers and 936 real examples (CHI 2025) shows that as trust in GenAI increases, the exercise of critical thinking decreases, changing nature — from production to verification, from execution to supervision. People also follow an AI’s advice more than a human expert’s, even against available contextual information, and this over-reliance is costly: in the experiment by Klingbeil, Grützner, and Schreck (Computers in Human Behavior, November 2024) those who followed the AI obtained lower payoffs than the control group. Those who delegate do not build competence nor the mental model to verify the result: immediate relief, long-term impoverishment — especially for young people, who build skills through AI, not before it.
Effect on the professions
- Interaction Designer — From task completion through navigation to goal delegation: the “Delegative UI” (Nielsen) becomes a key competence. Redesign around real intents (jobs-to-be-done), supporting multiple coexisting modes: guided path, quick command, browse-and-discover. → Useful resource: 18 Predictions for 2026 — Jakob Nielsen
- Product Manager — Identify in the roadmap which functions can be made agentic: from manual steps to delegable instructions. Discovery and comparison work in AI; the transaction is better kept where the user already has trust.
- Service Designer — Journeys with scenarios in which the user does not execute the flow but commissions it; the agent must be mapped as an autonomous actor. Responsibility in case of error and transparency on actions must be defined. The IBM Research study (Brachman et al., IUI ‘25) maps what users know and want to know about an agent — actions performed, tools, degree of confidence — and signals a specific obstacle: mental models built on other known systems interfere with understanding a new agentic system. Redesign journeys around outcomes, not touchpoints. → Useful resource: Building Appropriate Mental Models: What Users Know and Want to Know about an Agentic AI Chatbot (IUI ‘25)
- UX Researcher — How to study latent intents, not just observable behaviors? Clicks, paths, and heatmaps say what the user does, not what they want to achieve: it is necessary to strengthen effort in interviews on the desired outcome, ethnography, analysis of natural language requests.
- Information Architect — The architecture of intent is the prerequisite of delegation: an agent can only execute what the service exposes as an identifiable action. A function reachable only by traversing a menu hierarchy is not delegable.
- Content Strategist and UX Writer / Copywriter — The TL;DR becomes a primary element: opening summaries, titles that anticipate the conclusion, synthesis before detail — answers to a real need, not editorial shortcuts.