cognify.

Applied R&D · AI · Data · Analytics

AI-Native Product Design

AI-Native Product Design is a consultable library on how to redesign traditional product components with artificial intelligence capabilities. It is intended for UX/UI designers, product owners, and developers who need to decide when, how, and why to incorporate AI into an interface. Each pattern documents the conventional interaction, the AI-native variant, the usage criteria, the risks, and the control mechanisms.

Principles

  • Intent

    AI should operate on what the user wants to achieve, not only on what they type. Designing for intent means interpreting the goal behind an imprecise input and translating it into a structured action, without forcing the user to speak the system’s language.

  • Input flexibility

    Accept natural, fragmented, or incomplete input and normalize it without penalizing the user for not knowing the correct format. The interface absorbs variability; the user should not adapt to the system.

  • Evidence and uncertainty

    Every claim, result, or inference from the AI must show its evidence and its degree of certainty. Do not present an answer as fact when it is an estimate, and let the user inspect where each conclusion comes from.

  • Control and reversibility

    The user keeps agency: they can review what the AI interpreted, correct it, reject it, and undo it. Automation must never be irreversible without explicit consent.

  • Proportional use of AI

    AI is used when it adds real capability, not as decoration. If a conventional component solves the problem more simply, reliably, or cheaply, AI is not justified.

Pattern catalog

Six patterns on how to redesign traditional components with AI. The first is fully developed; the rest are exploratory.

  • Intent-based searchReinterpret search to operate on the user’s intent, not only on keywords.Developed
  • Flexible formsReplace rigid forms with natural-language input normalized into structured data.Exploratory
  • Interpretable tablesTables that explain, summarize, and answer questions about their own data.Exploratory
  • Contextual exportGenerate outputs adapted to context and destination from the content itself.Exploratory
  • Relevant notificationsNotifications that filter, prioritize, and draft by relevance to each user.Exploratory
  • Document loadingDocument intake that structures, classifies, and validates the loaded information.Exploratory

Fiche · Intent-based search

Conventional search matches query keywords against an index. Intent-based search interprets the user’s goal —what they want to find, filter, or decide— and translates that intent into a structured query, implicit filters, and ranking oriented to the expected outcome.

Conventional vs. AI-native interaction

Conventional

  • The user types exact terms.
  • They must know the system’s vocabulary (field names, categories, syntax).
  • Results depend on literal matching; synonyms or typos degrade them.
  • Filters and ordering are applied manually after the search.

AI-native

  • The user describes what they need in natural language.
  • The system interprets the intent and proposes filters, ordering, and scope.
  • It accepts ambiguous, incomplete, or colloquial queries.
  • It shows how it interpreted the query and lets the user correct it before running.

Usage criteria

  • The catalog or dataset is large and heterogeneous.
  • Users do not know the system’s exact vocabulary.
  • The user’s real intent is not reducible to keywords.
  • The cost of misinterpreting is low or reversible.

Risks

  • Overinterpretation: the system infers filters the user did not want.
  • Opacity: the user does not understand why they got those results.
  • Dependency on a model that can change behavior.
  • Silent degradation when the interpretation is wrong.

Control mechanisms

  • Show the interpretation as editable before or alongside the results.
  • Allow reverting to exact keyword search.
  • Expose inferred filters and let the user disable them.
  • Log the original query alongside the interpreted one.

Interactive demo — Intent-based search

SIMULATION

This demo does not use a real AI model. Interpretations are local deterministic rules that mimic, in a simplified way, how an intent-based search would reinterpret the query.

Other fiches (exploratory)

The following patterns are in an exploratory phase: we document the problem and the redesign hypothesis, without a developed fiche.

  • Flexible formsReplace rigid forms with natural-language input normalized into structured data, preserving validation and traceability.Exploratory
  • Interpretable tablesTables that explain, summarize, and answer questions about their own data, exposing evidence and uncertainty.Exploratory
  • Contextual exportGenerate reports, documents, or data adapted to context and destination from the content itself.Exploratory
  • Relevant notificationsNotifications that filter, prioritize, and draft by relevance to each user.Exploratory
  • Document loadingDocument intake that structures, classifies, and validates the loaded information, preserving traceability.Exploratory