<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Han-Shen Chen on Medium]]></title>
        <description><![CDATA[Stories by Han-Shen Chen on Medium]]></description>
        <link>https://medium.com/@han_AIUX?source=rss-90d8324e18fb------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*Tp0kcQ8PxI_l-wzn8de8ng@2x.jpeg</url>
            <title>Stories by Han-Shen Chen on Medium</title>
            <link>https://medium.com/@han_AIUX?source=rss-90d8324e18fb------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 05:04:28 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@han_AIUX/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[AI Agentic Design — Part 1]]></title>
            <link>https://medium.com/@han_AIUX/ai-agentic-design-part-1-dd25b73e0912?source=rss-90d8324e18fb------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd25b73e0912</guid>
            <dc:creator><![CDATA[Han-Shen Chen]]></dc:creator>
            <pubDate>Fri, 18 Apr 2025 00:44:03 GMT</pubDate>
            <atom:updated>2025-04-18T01:17:56.444Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI Agentic Design — Part 1</h3><p>From Input/Output to Intent/Outcome: A GenAI UX Design Mindset</p><h3>The Traditional Model: Input / Output</h3><p>Traditionally, the interaction model between users and computer systems has been framed around a simple process: the user provides an <strong>input</strong>, and the system returns an <strong>output</strong>. This concept has formed the basis of interaction design for decades. Buttons, forms, commands — all are mechanisms that rely on users giving something to the system so it can respond appropriately.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HSIUef5ewtYkaVy54kBxvw.png" /></figure><p>But in reality, this model is based on an illusion of control. Designers often already know what the system can output. As a result, most of UX design has focused on guiding users toward providing the <em>correct input</em> — one that aligns with the system’s expectations and limitations. In essence, we design the input experience based on the output we <em>already understand</em>.</p><p>This reveals a fundamental paradox: what appears to be a user-driven process is actually output-driven design. UX becomes the bridge to ensure the user’s input fits what the system can handle. While this model has served us well in traditional software design, it begins to show its limitations in the age of generative AI.</p><h3>The Shift: Intent / Outcome in the GenAI Era</h3><p>With the rise of GenAI systems, the user-AI relationship has fundamentally evolved. Today’s AI models are capable of interpreting complex, ambiguous, and multimodal <strong>intents</strong> — whether expressed through text, speech, visuals, or actions. Rather than expecting a perfectly formatted input, AI systems can understand <em>what the user wants</em>, even if it’s not stated precisely.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rsi5EaWZ4kFuFitw91OfwA.png" /></figure><p>At the center of this new interaction model is the <strong>GenAI agent</strong>. These agents don’t just relay inputs to a system — they interpret the user’s <strong>intent</strong>, convert it into structured <strong>input</strong>, and then orchestrate responses across multiple AI tools, including large language models (LLMs), vision models, and external APIs. The agent functions as a real-time interpreter and operator.</p><h3>Agents as Orchestrators of Multimodal Outcomes</h3><p>What comes back from the AI system is no longer just a raw output. The GenAI agent now takes on the critical task of transforming those results into meaningful <strong>outcomes</strong>. These outcomes are tailored to the user’s intent and preferred modality. For instance:</p><ul><li>A request could yield a <strong>voice response</strong> with <strong>visual explanations</strong></li><li>A user might receive an <strong>image summary</strong> alongside <strong>actionable text</strong></li><li>A video might be analyzed and returned as <strong>narrated scene breakdowns</strong></li></ul><p>In this model, the agent not only enables human-to-machine interaction but also synthesizes machine-to-human communication in rich, contextual, and intuitive ways.</p><h3>A New UX Loop: Intent → Input → Output → Outcome</h3><p>This shift marks a new paradigm for UX design. No longer confined to optimizing inputs for known outputs, UX in the GenAI era becomes about designing experiences that surface <strong>intent</strong> and enable <strong>outcomes</strong> — even when neither is precisely defined from the start.</p><p>The GenAI agent ensures:</p><ul><li>Users don’t need to phrase things perfectly</li><li>The system understands nuance and multimodal signals</li><li>The outcome is personalized, contextual, and aligned with the user’s goals</li></ul><h3>Conclusion: Evolving the UX Mindset</h3><p>As we move from Input/Output to Intent/Outcome, the UX design mindset must evolve. It’s no longer just about usability — it’s about <strong>interpretability</strong>, <strong>orchestration</strong>, and <strong>alignment</strong>. GenAI agents empower this new loop, transforming how we define interaction, usability, and experience.</p><p>In this new world, UX is no longer the craft of optimizing the <em>input</em> field — it’s the art of aligning human intent with system capability to deliver meaningful, multimodal <em>outcomes</em></p><h3>從輸入/輸出到意圖/結果：GenAI UX 設計新思維</h3><h3>傳統模型：輸入 / 輸出</h3><p>在過去的互動設計中，使用者與電腦系統的關係被簡化為一個單純的過程：使用者提供<strong>輸入</strong>，系統則回傳<strong>輸出</strong>。這套模型長期以來是我們設計互動介面的基本邏輯，從按鈕、表單、指令到各式各樣的操作方式，都是以這個概念為基礎。</p><p>但實際上，這個模型並不完全正確。設計師早就知道系統會給出什麼樣的輸出，因此整個 UX 設計的目的，其實就是引導使用者提供<strong>正確的輸入</strong>。換句話說，我們設計「輸入」的方式，是根據我們已知的「輸出」來反推的。</p><p>這揭示了一個本質上的矛盾：表面上看似是由使用者驅動的互動，其實背後是以系統輸出為主導的設計邏輯。雖然這個框架在傳統軟體設計中行之有效，但在生成式 AI 的時代，它的限制也越來越明顯。</p><h3>轉變：在 GenAI 時代下的意圖 / 結果模型</h3><p>生成式 AI 的出現徹底改變了人與系統之間的互動模式。現今的 AI 模型已經能夠理解複雜、模糊，甚至是多模態的<strong>意圖</strong> — — 無論是文字、語音、圖像還是行為表現。使用者不再需要提供完美格式化的輸入，AI 系統可以從中理解出使用者真正的需求。</p><p>而在這個互動模型的中心，就是<strong>GenAI 代理（Agent）</strong>。這些代理不再只是轉發使用者輸入給系統，而是主動<strong>解析使用者的意圖</strong>，轉換成有結構的<strong>輸入</strong>，並在背後協調大型語言模型（LLM）、影像模型或外部 API 等 AI 工具完成任務。代理的角色，就像一位即時的解譯員與操作員。</p><h3>GenAI 代理：多模態結果的協調者</h3><p>從 AI 系統取得的資訊，也不再是單一的「輸出」。GenAI 代理接著擔任整合與轉譯的角色，將系統回傳的結果，整理成對使用者而言有意義的<strong>結果（Outcome）</strong>，並依據使用者的偏好與語境以最佳方式呈現。</p><p>這些結果可能包括：</p><ul><li>視覺化圖表搭配自然語言解釋</li><li>音訊摘要加上文字行動建議</li><li>影片場景的分段說明與敘述</li><li>來自多模型推理的具體任務回應</li></ul><p>在這種模式下，GenAI 代理不僅支援人與機器的互動，也協助將機器的能力，以直覺、具語境、個人化的方式回饋給人。</p><h3>全新的 UX 互動循環：意圖 → 輸入 → 輸出 → 結果</h3><p>這代表 UX 設計邁向一種全新的設計思維。我們不再只是優化使用者的「輸入」操作，而是設計一整個流程，幫助使用者表達<strong>意圖</strong>，並接收個人化的<strong>結果</strong>。即使使用者無法明確描述需求，GenAI 系統仍能透過代理理解與轉譯。</p><p>這樣的 UX 互動循環，使得：</p><ul><li>使用者無需說得精準，系統也能理解</li><li>系統能解析語意、情境與多模態訊號</li><li>回應結果與使用者目標一致，形式多元、語境貼近</li></ul><h3>結語：UX 心態的轉變</h3><p>當我們從「輸入/輸出」走向「意圖/結果」，UX 設計的心態也必須進化。這不再僅僅是使用性與易用性的問題，而是關於<strong>理解力（Interpretability）</strong>、<strong>協調力（Orchestration）與一致性（Alignment）</strong>。</p><p>GenAI 代理是這場轉變的關鍵，使我們得以重新定義人機互動的方式。在這個新世界中，UX 不再只是幫助使用者「點對的按鈕」，而是讓系統能夠理解人的需求，並提供能夠真正實現目標的<strong>多模態結果體驗</strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd25b73e0912" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>