AI Patterns / AI Interaction Pattern 12
Human-Agent Collaboration Interfaces
Design interfaces for directing, monitoring, and providing feedback to proactive "Agentic AI" that can automate complex, multi-step workflows.
The User Problem This Pattern Solves
Many valuable business workflows—like conducting market research, analyzing sales data, or preparing a comprehensive report—are too complex for a simple, single-shot AI command. They require multiple sequential steps, such as finding data, cleaning it, analyzing it, and then synthesizing the results into a new format. Users need a way to delegate these complex goals to an AI and trust that it will execute them correctly.
The Design Solution & UI Mockup
The solution is an interface that reframes the user's relationship with the AI from a simple Q&A to a manager-assistant dynamic. The user provides a high-level goal. The AI, acting as an agent, responds not with an answer, but with a *plan*—a clear, step-by-step breakdown of the actions it will take. The user can then review, approve, and monitor the plan's execution in real-time. This "glass box" approach builds trust and gives the user ultimate supervisory control over the autonomous agent.
Agent's Execution Plan:
-
✅
Accessing `sales_data_q4.csv` from SharePoint.
-
⏳
Analyzing sales data by region and product line.
-
...
Identifying top 3 performing products and key trends.
-
...
Generating summary charts and graphs.
-
...
Creating draft presentation `Q4_Sales_Summary.pptx`.
Key Benefits & Impact
Automates Complex Workflows
Goes beyond simple tasks to automate entire multi-step business processes.
Acts as a Force Multiplier
Allows a single user to manage multiple complex tasks simultaneously, dramatically increasing their output.
Builds Trust in Autonomy
The transparent plan and real-time monitoring give users the confidence to delegate high-stakes tasks to an AI.
Design Considerations
The ability to interrupt and redirect the agent is non-negotiable. The user must always have an easily accessible "stop" button. The interface should also provide clear error handling; if a step fails, the agent should pause, report the error, and ask the user for guidance on how to proceed. This ensures the human always remains the ultimate decision-maker in the collaboration.
Capabilities →
All Work →
Dashboard Design System →
AI Interaction Patterns →
About →
Skills →