---
title: "Comment by Tarun Sharma on \"Agentic UX: Designing Interfaces for Autonomous AI Decision-Making\" | Hashnode"
description: "The agent-state-as-first-class-concept point maps almost exactly onto a state machine, planning, executing, waiting-for-approval, completed, needs-intervention"
url: "https://hashnode.com/posts/agentic-ux-designing-interfaces-for-autonomous-ai-decision-making/6aa07beb4404e516b876907c/comment/6aa46c70528108efad5cfe85"
---

[Discussion](https://hashnode.com/posts/agentic-ux-designing-interfaces-for-autonomous-ai-decision-making/6aa07beb4404e516b876907c)

Comment [Agentic UX: Designing Interfaces for Autonomous AI Decision-Making](https://hashnode.com/posts/agentic-ux-designing-interfaces-for-autonomous-ai-decision-making/6aa07beb4404e516b876907c)

[Tarun Sharma](https://hashnode.com/@101beardo)

[Sep 11](https://hashnode.com/posts/agentic-ux-designing-interfaces-for-autonomous-ai-decision-making/6aa07beb4404e516b876907c/comment/6aa46c70528108efad5cfe85)

The agent-state-as-first-class-concept point maps almost exactly onto a state machine, planning, executing, waiting-for-approval, completed, needs-intervention are states with explicit valid transitions, not just a status string you branch on. Something like XState models this well since it makes illegal transitions (like completing from idle without executing) impossible by construction, instead of relying on every consumer of the state to check it correctly.

[Kalpick Sharma](https://hashnode.com/@kalpicksharma)

[Sep 12](https://hashnode.com/posts/agentic-ux-designing-interfaces-for-autonomous-ai-decision-making/6aa07beb4404e516b876907c/comment/6aa46c70528108efad5cfe85#reply-6aa4a191db647b42e4759326)

Great point — you're right. Agent state should be a state machine with explicit valid transitions, not a string to branch on. XState-style enforcement (illegal transitions impossible by construction) is the better model. Will update the article. Thanks!