Agentic UX: Designing the "Mind"

For the last 15 modules, we have built amazing technical brains. But if your user sees a simple text box and a "loading" spinner for 30 seconds, they will hate your product. Agentic UX is about making the agent's complex reasoning visible, interactive, and trustworthy.

1. The "Black Box" Problem

In a standard ChatGPT interface, you type a prompt and wait. For an agentic system that is searching the web, analyzing data, and writing code, this "Empty Wait" creates anxiety.

• "Is it broken?"
• "What is it doing with my data?"
• "Why is it taking so long?"

---

2. The Solution: Radical Transparency

Instead of a spinner, show the Agent's Plan.

• Searching for stock prices...
• Calculating P/E ratio...
• Drafting final report... This "Progressive Disclosure" of thoughts makes the latency feel like "Work" rather than "Lag."

---

3. Visualizing the UX Shift

UI Pattern	Static AI (Chat)	Agentic AI
Input	Single Prompt	Goal + Constraints
Status	Pulsing Dot	Action Step Logs
Output	Paragraph of Text	Live-updating Dashboard
Interaction	Response only	Intervention (Edit the plan)

---

4. The "Canvas" vs. The "Thread"

New AI tools (like Claude Artifacts or ChatGPT Canvas) move away from a vertical thread.

• The Thread is for talking.
• The Canvas is for building. When an agent is creating code or a report, it should "Pop out" into a side-panel where the user can see it grow in real-time.

---

5. Trust through "Traceability"

Every claim the agent makes should be Interactive.

• If the agent says "Sales are up 20%," the user should be able to click that number and see the Source Tool Result (the raw SQL or PDF snippet) that generated it.
• The Rule: Never show a fact without a "Why."

---

Key Takeaways

• Chat boxes are a low-bandwidth way to interact with a high-bandwidth agent.
• Visibility of the agent's plan reduces user anxiety and builds trust.
• Progressive Disclosure turns wait time into information.
• Traceability allows users to verify AI work without reading logs.