NLX Design and Agent Framework

In the AI era, Natural Language Interface (NLX) is the new UX. While traditional GUIs are rigid, NLX is elastic. However, a conversational interface is not a "no-design" zone; it requires a new set of invisible grammars, structures, and UI constructs to prevent the creation of "Frankenstein products."

1. Defining the Agent (The 3-Part Test)

When building an AI agent rather than a simple assistant, evaluate it against these three criteria:

• Autonomy: Move from "human-in-the-driver's-seat" to delegation. It should operate on a spectrum of independence where you provide a goal, and the agent determines the path.
• Complexity: The task must be multi-step and non-linear. If it is a "one-shot" task (e.g., "summarize this"), it is an assistant. If it is "build a prototype for this idea," it is an agent.
• Asynchronous Nature: The agent should work when the user is not working. It should be able to join meetings, perform deep research, or execute background processes without constant human monitoring.

2. The NLX Design Construct

Treat natural language as a designed interface using these specific elements:

• The Prompt as the New PRD: Use prompts to define the product’s logic and constraints. If you cannot prototype the behavior via a prompt set, the product is not yet defined.
• Editable Plans: When a user provides a high-level goal, the agent should return a "Plan" construct. This plan must be visible and editable by the user before execution to build trust.
• Showing the Work (Thinking Aloud): Design the "inference time" experience.
  • Too verbose: Feels like a technical script/cron job.
  • Too terse: Creates a "black box" that reduces user confidence.
  • The Sweet Spot: Provide enough breadcrumbs to show the agent is on the right path without overwhelming the user.
• Proactive Follow-ups: Design the "happy path" by suggesting the next logical steps (e.g., if an image is generated, suggest "Make this high-resolution" or "Change the style to watercolor").