12th March 2025, Kathmandu
Artificial Intelligence (AI) agents are rapidly evolving, promising to revolutionize industries by automating complex tasks and workflows. However, there’s a glaring issue that no one is talking about: the user interface (UI) for AI agents is broken.
Missing UI For AI Agents
While the tech world is obsessed with making AI agents autonomous, almost no one is thinking about how humans will interact with them. As someone who started their career as a UX engineer, I’m here to sound the alarm: AI agents need better UIs, not just better prompts.
The Problem with Chatbots as AI Interfaces
Right now, most AI agents are glorified chatbots. While chatbots are great for simple interactions, they fall short when it comes to complex workflows and collaboration. Here’s why:
Complex Workflows: Chat interfaces struggle to visualize dependencies, data, or progress. Users are left scrolling through endless text logs, trying to piece together what’s happening.
Edits & Reversibility: Fixing AI mistakes via chat is clunky and inefficient. There’s no easy way to undo or adjust actions without starting over.
Multi-agent collaboration: When multiple AI agents and humans need to work together, chat interfaces become chaotic and disorganized.
Cognitive Load: Endless chat logs overwhelm users, making it hard to focus on the task at hand.
If AI agents are going to handle real-world tasks, their UIs need to reflect the complexity and nuance of the work being done. So, what’s the solution?
The Missing UI for AI Agents: 4 Key Innovations
To unlock the full potential of AI agents, we need to move beyond chatbots and develop UIs that are intuitive, collaborative, and visually engaging. Here are four groundbreaking approaches to the UI for AI agents:
1. Agent Dashboards (Observable Autonomy)
Imagine a command center where you can monitor and guide AI decision-making in real time. Agent dashboards would:
Display the AI’s reasoning process and confidence levels.
Allow users to step in at critical decision points without micromanaging.
Provide a clear overview of tasks, progress, and outcomes.
Example: An AI assistant that schedules meetings but lets you visually adjust priorities and timelines.
2. Workflow Graphs (Dynamic Task Maps)
Workflow graphs would represent tasks, decisions, and AI actions as editable visual maps. This approach would:
Make multi-agent collaboration spatially organized and easy to follow.
Allow users to modify workflows visually instead of typing commands.
Provide a bird’s-eye view of complex processes.
Example: An AI planning a marketing campaign as an editable graph instead of a text-based list.
3. Editable AI Notebooks (Actionable Memory)
Instead of ephemeral chat logs, AI agents could maintain structured, editable documents that evolve. These notebooks would:
Serve as a persistent knowledge base that users can refine and reference.
Allow AI to continuously update and improve its outputs.
Make workflows traceable and reusable.
Example: An AI investment research agent that builds a due diligence report as a living document.
4. Agent Multimodal UIs (Beyond Text)
The future of AI interaction lies beyond text. Multimodal UIs would enable users to interact with AI agents through:
Drag-and-drop interfaces.
Voice commands.
Spatial and gesture-based controls.
Example: A robotics control UI where human operators guide AI via gestures, similar to Medivis’s innovative interfaces.
Why UI Matters for AI Agents
The current focus on autonomy and prompt engineering is only half the battle. To truly integrate AI into our workflows, we need UIs that bridge the gap between human intuition and machine efficiency. Here’s why UI is the missing piece of the puzzle:
Observability: Users need to see what’s happening inside the machine. This includes understanding inputs, variables, and reasoning processes.
Data Visibility: When multiple APIs or agents interact, users need a clear view of the data flow and decision-making process.
Multi-Threading: AI agents must handle multiple tasks simultaneously, and the UI should reflect this complexity without overwhelming the user.
The Big Idea: UI Defines the Next Era of Computing
The AI world is currently stuck in the “single-threaded chatbot” phase. But the real opportunity lies in building UIs that enable seamless collaboration between humans and AI at scale. Whoever cracks the UI for agentic software will define the next era of computing.
Who’s Working on This?
Surprisingly, almost no one is tackling this challenge head-on. While companies are racing to build smarter AI agents, few are investing in the interfaces that will make these agents usable and effective in real-world scenarios. This gap represents a massive opportunity for innovators and entrepreneurs.
What Do You Think?
Which of these UI approaches resonates most with you? Are there other ideas or examples you’ve seen that could revolutionize how we interact with AI agents? Let’s start a conversation about the future of AI interfaces and how we can make them work for humans, not just machines.
By rethinking the UI for AI agents, we can unlock their full potential and create tools that truly enhance human productivity and creativity. Let’s stop settling for chatbots and start building the interfaces of the future.
For more: Missing UI For AI Agents