08 August 2025
These are the design principles from PAIR, a research team from Google, for designing AI empowered products.
Source: People + AI Guidebook
AI-supported workflows should be designed for the appropriate level of user autonomy, that consider different user tasks, expertise, and the overall effort required to steer the AI system.
People expect tools and technologies to improve their efficiency and productivity. People also prefer to remain in control of which tasks tools should execute, how, and to what end. Knowing when a user is willing to delegate a task to an AI, how users can control the AI, and when to leave control with the user entirely is key for building helpful AI products.
People engage differently with AI systems based on their context. Align datasets and models to people’s interactions with AI systems in real world contexts.
As your AI products scales, it may be utilized for unexpected tasks, socio-technical contexts, or knowledge domains. Natural and intuitive interactions with AI systems can differ in surprising ways from standard benchmarks, training or tuning data. It’s essential to align your user journeys, data collection and model evaluation efforts with AI interactions that might occur in real-world contexts, as early as possible.
AI products should have a multifaceted safety strategy that evolves with technology and people. Account for varying degrees of risk and ensure people’s safety, now and in the future.
As your AI product gains traction, users will perform more intricate and complex tasks. Changes in user context and complexity can introduce new or exacerbate existing risks, beyond the scope of your AI product. Proactively develop safeguards for different risk levels in your product’s user experience and AI models. Be prepared to adapt and modify safety strategies to respond to evolving trends and behaviors.
Adapt your AI system with feedback from people, during individual interactions and over time.
Human-AI interactions are a bidirectional feedback loop. AI learns from users to personalize their experiences, and users adapt their behaviors and workflows in response to AI outcomes. Set up feedback mechanisms that can be used to interpret AI outcomes, and account for changes introduced by AI.
Create helpful AI experiences that enhance aspects of work & play that people enjoy and inspire creativity.
AI can amplify human expertise, but the efficiency gains aren’t always assured, cheaper, or enjoyable. People have certain expectations of benefit and performance that AI should deliver. Focus on creating experiences that seamlessly integrate and enhances aspects of work and play that people like—and even love, in their workflows.
AI-supported workflows should be designed for the appropriate level of user autonomy, that consider different user tasks, expertise, and the overall effort required to steer the AI system.
People expect tools and technologies to improve their efficiency and productivity. People also prefer to remain in control of which tasks tools should execute, how, and to what end. Knowing when a user is willing to delegate a task to an AI, how users can control the AI, and when to leave control with the user entirely is key for building helpful AI products.
People engage differently with AI systems based on their context. Align datasets and models to people’s interactions with AI systems in real world contexts.
As your AI products scales, it may be utilized for unexpected tasks, socio-technical contexts, or knowledge domains. Natural and intuitive interactions with AI systems can differ in surprising ways from standard benchmarks, training or tuning data. It’s essential to align your user journeys, data collection and model evaluation efforts with AI interactions that might occur in real-world contexts, as early as possible.
AI products should have a multifaceted safety strategy that evolves with technology and people. Account for varying degrees of risk and ensure people’s safety, now and in the future.
As your AI product gains traction, users will perform more intricate and complex tasks. Changes in user context and complexity can introduce new or exacerbate existing risks, beyond the scope of your AI product. Proactively develop safeguards for different risk levels in your product’s user experience and AI models. Be prepared to adapt and modify safety strategies to respond to evolving trends and behaviors.
Adapt your AI system with feedback from people, during individual interactions and over time.
Human-AI interactions are a bidirectional feedback loop. AI learns from users to personalize their experiences, and users adapt their behaviors and workflows in response to AI outcomes. Set up feedback mechanisms that can be used to interpret AI outcomes, and account for changes introduced by AI.
Create helpful AI experiences that enhance aspects of work & play that people enjoy and inspire creativity.
AI can amplify human expertise, but the efficiency gains aren’t always assured, cheaper, or enjoyable. People have certain expectations of benefit and performance that AI should deliver. Focus on creating experiences that seamlessly integrate and enhances aspects of work and play that people like—and even love, in their workflows.
