Gen AI Assistant
Product design
Company
Apple
Role
Lead Product Designer
Duration
Feb 2025 - Jul 2025
This project features an in-product Gen AI Assistant that transforms how users interact with their data—using natural language to uncover insights and automatically generate workflows that connect third-party apps and eliminate manual work.
Let me help you with that.
Users struggle to make sense of their data and spend countless hours on repetitive tasks that could be automated, but lack the technical expertise to build cross-platform workflows or query complex datasets effectively.
Problem
Build an intelligent assistant that bridges the gap between user intent and technical execution, making data analysis and cross-platform automation accessible through natural language conversation.
Opportunity
The Challenges
While the concept seemed straightforward—let users talk to their data and automate workflows—the execution revealed several complex design challenges that required careful consideration of user trust, technical constraints, and interaction patterns.
Challenge 1: Managing user mental models
One concern was that users would approach our AI assistant with conflicting expectations shaped by existing tools, creating design tension between familiarity and innovation.
Competing mental models
ChatGPT/Claude users
Would expect pure conversational text responses, likely feel confused by visual workflow editors.
Traditional automation users
Would anticipate manual step-by-step configuration through forms and dropdowns, might not understand natural language workflow requests.
Business intelligence users
Would expect traditional dashboard interfaces for data analysis, potentially unsure how to conversationally explore data.
Our AI combined conversational AI, workflow automation, and data analysis in ways that didn't exist in users' reference points. We needed to leverage familiar patterns to reduce learning curve while introducing new capabilities that users hadn't experienced before.
How do you create an interface that feels approachable through familiarity, yet distinct enough to communicate unique AI-powered value?
Iteration
Users start with a simple, open-ended question paired with natural language examples that immediately demonstrate the AI's capabilities, making the tool approachable while setting expectations for conversational interaction.
After the initial prompt, familiar “+” symbols in both panels invite the user to interact with the proposed solution, helping the user discover that the tool is both conversational and interactive.
If users skip the initial prompt, text in the blank area instructs the user to add objects manually or chat with the AI to start building their solution, helping the user discover that the tool is both conversational and interactive.
Solution
We designed a unified interface that strategically borrowed familiar elements from each mental model while introducing users to new AI capabilities through progressive exposure.
Leveraging familiar patterns
Chat-first entry point
Users start with the familiar conversational interface they know from ChatGPT/Claude, making the tool immediately approachable.
Visual workflow representation
Once AI generates automation, we present it in a familiar layout that resembles traditional automation tools users recognize.
Introducing new concepts gradually
AI as workflow generator
Users discover through conversation that the AI can create entire automations, not just answer questions.
Conversational refinement
After seeing the visual workflow, users learn they can continue chatting to modify it rather than clicking through forms.
Progressive interactivity discovery
Invite exploration; use familiar call to actions that once users click one and discover they can add workflow steps, they realize the entire flow panel is interactive and editable.
Challenge 2: Balancing automation power with user control
Users need powerful AI automation but also require control to trust and modify these processes, creating tension between efficiency and user agency. We found these three key points:
Trust varies by context
Users comfortable with AI pulling sales data may want manual control for customer-facing actions like sending emails or modifying CRM records.
Different comfort levels
Some users want full automation with just results, while others need to see and approve every step.
Business-critical sensitivity
Higher stakes with third-party integrations and sensitive data require more transparency and control.
So how do we create an experience that feels effortless for users who want maximum automation, while providing the transparency and control others need to feel confident?
Iteration
To accompany the chat interface, we designed a flexible three-panel workspace that gives users multiple ways to interact with their AI-generated content:
Preview panel
Visualizes data and app screen flows
Objects Panel
Provides drag-and-drop components to add to the Preview
Flows Panel
Shows the automation workflow connecting third-party apps
The Objects panel intelligently adapts based on the user's current context. When viewing the Flows panel, it automatically shifts to display available third-party app connectors that users can drag directly into their automation workflow, making it easy to extend or modify AI-generated automations.
This approach builds trust through transparency while preserving efficiency and user preference. Users get the speed of automation for workflows the AI generates correctly, but retain complete control to refine them either through direct manipulation or continued conversation—whatever feels more natural for their working style.
Solution
We designed a hybrid approach where the AI generates complete workflows but presents them as editable, visual flows that users can inspect and modify.
Key insight: Treat AI-generated workflows as intelligent first drafts, not final products.
Full transparency
Users see exactly what the AI created: which apps connected, data sources, and information flow between steps.
Dual interaction modes
Users choose their preferred modification method.
Manual editing: directly add, delete, or change steps in the visual interface
Conversational refinement: continue using natural language to request AI modifications
Best of both worlds
Speed of automation for workflows the AI gets right, plus complete control to refine them using whatever method feels most natural.
The Final Design
After wrestling with competing mental models, balancing automation with agency, and figuring out how to make AI feel collaborative rather than mysterious, we arrived at designs we're proud of. This AI assistant transforms how users interact with their data and workflows—making complex automation feel conversational and putting users firmly in control of their AI-generated solutions.
01
Starting Simple
Approachable
We designed the starting point to feel like a conversation, not a form. The open question paired with practical examples shows users the breadth of what they can create while establishing that natural language is the primary way to interact with the system.
02
Three Panel View
Sectioned and Organized
The three-panel workspace creates clear separation between conversation, visualization, and automation editing, allowing users to focus on one aspect at a time while maintaining awareness of the complete workflow. This organization prevents cognitive overload by giving each interaction mode its own dedicated space while keeping everything contextually connected.
Detailed View
The generous space given the combined Preview and Flow panels invites users to click into objects within each panel for a detailed view, helping them discover the dynamic, interactive nature of the AI tool and allowing them to verify details of the proposed solution.
03
Manual Flexibility
Add Object
We added the ability for users to manually add or move objects around in the AI-generated solution. This allows users the flexibility to make their own adjustments if they can't put the changes they want into words, or if they simply want to take matters into their own hands.
Conditionals
The blue underlined conditional logic shows users they can click to modify the "if-then" rules governing their automation, providing granular control over when and how workflows execute without needing to understand complex programming syntax.
Third-party App Control
Connecting to third-party apps requires a lot of trust, so we wanted to provide users manual control to make adjustments to the actions and accounts associated with the third-party apps in the automation flow.
Complete Integration Options
When manually adding steps to their workflow, users have access to the full library of available connectors and conditionals, ensuring they can customize automations beyond the AI's initial suggestions. This comprehensive access means users aren't limited by what the AI chose to include and can build exactly the workflow they envision.
