<?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 Itamar Medeiros on Medium]]></title>
        <description><![CDATA[Stories by Itamar Medeiros on Medium]]></description>
        <link>https://medium.com/@designative?source=rss-6c2b386e05b2------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*5qIPc1jOH9GAu0aWqf0kHg.jpeg</url>
            <title>Stories by Itamar Medeiros on Medium</title>
            <link>https://medium.com/@designative?source=rss-6c2b386e05b2------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 17:13:08 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@designative/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[From Intent to Alignment: Why Design Leadership Must Institutionalize Intent Documentation]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/from-intent-to-alignment-why-design-leadership-must-institutionalize-intent-documentation-eab9234344bd?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1600/0*r6acEO-axArRfEs6.jpg" width="1600"></a></p><p class="medium-feed-snippet">Agentic AI challenges traditional requirements; intent documentation aligns teams and guides Agents beyond the UI.</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/from-intent-to-alignment-why-design-leadership-must-institutionalize-intent-documentation-eab9234344bd?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/from-intent-to-alignment-why-design-leadership-must-institutionalize-intent-documentation-eab9234344bd?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/eab9234344bd</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[design-management]]></category>
            <category><![CDATA[design-strategy]]></category>
            <category><![CDATA[user-experience-design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Tue, 12 May 2026 22:11:07 GMT</pubDate>
            <atom:updated>2026-05-12T22:11:07.989Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Designing for Explainability: From Interrogability to Intent Alignment]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/designing-for-explainability-from-interrogability-to-intent-alignment-ef2af2c5019f?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*9LheBfcv_uDP_52h" width="1200"></a></p><p class="medium-feed-snippet">When agents act on our behalf, trust depends on more than answers. It depends on explainability so that the agent &#x201C;show the work&#x201D;.</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/designing-for-explainability-from-interrogability-to-intent-alignment-ef2af2c5019f?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/designing-for-explainability-from-interrogability-to-intent-alignment-ef2af2c5019f?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/ef2af2c5019f</guid>
            <category><![CDATA[ux-design]]></category>
            <category><![CDATA[ux-for-ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[explainable-ai]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Mon, 11 May 2026 08:09:42 GMT</pubDate>
            <atom:updated>2026-05-11T08:09:42.049Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Dual Evaluation: Measuring Both Human Experience and Agent Effectiveness]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/dual-evaluation-measuring-both-human-experience-and-agent-effectiveness-61c6b67102b0?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*OyVohrnwdunRhHsV" width="1200"></a></p><p class="medium-feed-snippet">H-ACD&#x2019;s &#x201C;Dual Evaluation&#x201D; principle explains why great agentic experiences succeed only when both humans and agents perform well together.</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/dual-evaluation-measuring-both-human-experience-and-agent-effectiveness-61c6b67102b0?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/dual-evaluation-measuring-both-human-experience-and-agent-effectiveness-61c6b67102b0?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/61c6b67102b0</guid>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Sun, 10 May 2026 10:08:08 GMT</pubDate>
            <atom:updated>2026-05-10T10:08:08.476Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Usability in the Age of AI]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/usability-in-the-age-of-ai-60e08c9b3552?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*MDarfEeqQuUPQ2Qg" width="1200"></a></p><p class="medium-feed-snippet">Traditional usability metrics fail for AI agents. Learn what changes when users delegate instead of navigate &#x2014; and what stays the same.</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/usability-in-the-age-of-ai-60e08c9b3552?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/usability-in-the-age-of-ai-60e08c9b3552?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/60e08c9b3552</guid>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Sat, 09 May 2026 21:08:35 GMT</pubDate>
            <atom:updated>2026-05-09T21:08:35.155Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Beyond the Conversation Trap: Designing for Hybrid Human-Agent Interaction Modes]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/beyond-the-conversation-trap-designing-for-hybrid-human-agent-interaction-modes-4e4d2e391ad9?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*tJjw3r6rDZv-LcKy" width="1200"></a></p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/beyond-the-conversation-trap-designing-for-hybrid-human-agent-interaction-modes-4e4d2e391ad9?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/beyond-the-conversation-trap-designing-for-hybrid-human-agent-interaction-modes-4e4d2e391ad9?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e4d2e391ad9</guid>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Sat, 09 May 2026 09:47:31 GMT</pubDate>
            <atom:updated>2026-05-09T18:20:21.375Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Preventing Agent Drift: Designing AI Systems That Stay Aligned With Human Intent]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/preventing-agent-drift-designing-ai-systems-that-stay-aligned-with-human-intent-40295c403766?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*N8NJ1uP7830vwCy7" width="1200"></a></p><p class="medium-feed-snippet">What would it take for AI agents to truly support human work &#x2014; capturing intent, preserving context, and staying aligned with real&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/preventing-agent-drift-designing-ai-systems-that-stay-aligned-with-human-intent-40295c403766?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/preventing-agent-drift-designing-ai-systems-that-stay-aligned-with-human-intent-40295c403766?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/40295c403766</guid>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[artficial-intelligence]]></category>
            <category><![CDATA[utterance]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Fri, 08 May 2026 11:10:11 GMT</pubDate>
            <atom:updated>2026-05-08T11:10:11.674Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Operationalizing AI Ethics in Design: From High-Level Principles to Encapsulated Workflows]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/operationalizing-ai-ethics-in-design-from-high-level-principles-to-encapsulated-workflows-14b872342327?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*yefJBbFgHb_WF7uB" width="1200"></a></p><p class="medium-feed-snippet">How can designers operationalize AI ethics? Turn ethical principles into workflow guardrails that make agentic systems safe, fair, and&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/operationalizing-ai-ethics-in-design-from-high-level-principles-to-encapsulated-workflows-14b872342327?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/operationalizing-ai-ethics-in-design-from-high-level-principles-to-encapsulated-workflows-14b872342327?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/14b872342327</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[design-strategy]]></category>
            <category><![CDATA[ethics]]></category>
            <category><![CDATA[ai-ethics]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Fri, 08 May 2026 08:00:58 GMT</pubDate>
            <atom:updated>2026-05-08T08:00:58.073Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Collaboration in the Age of AI: From groupware to Agent-Mediated Teamwork]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/collaboration-in-the-age-of-ai-from-groupware-to-agent-mediated-teamwork-f9992861c4a3?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1536/0*LFJ57c_hdyqykdEU.jpg" width="1536"></a></p><p class="medium-feed-snippet">To create AI tools to support collaboration, Agents must move from supporting individuals to shaping team alignment, trust, and&#x2026;</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/collaboration-in-the-age-of-ai-from-groupware-to-agent-mediated-teamwork-f9992861c4a3?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/collaboration-in-the-age-of-ai-from-groupware-to-agent-mediated-teamwork-f9992861c4a3?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/f9992861c4a3</guid>
            <category><![CDATA[teamwork]]></category>
            <category><![CDATA[agentic-ai]]></category>
            <category><![CDATA[collaboration]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[cscw]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Tue, 24 Feb 2026 16:00:27 GMT</pubDate>
            <atom:updated>2026-05-13T18:54:34.981Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[Workflow Encapsulation in Human–Agent Centered Design: From Doing Work to Governing Outcomes]]></title>
            <link>https://medium.com/design-bootcamp/workflow-encapsulation-in-human-agent-centered-design-from-doing-work-to-governing-outcomes-de416b27226d?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/de416b27226d</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[design-strategy]]></category>
            <category><![CDATA[ethics-in-tech]]></category>
            <category><![CDATA[jobs-to-be-done]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Mon, 02 Feb 2026 10:47:25 GMT</pubDate>
            <atom:updated>2026-02-02T10:47:25.786Z</atom:updated>
            <content:encoded><![CDATA[<p>In software engineering, encapsulation traditionally meant hiding implementation details behind stable interfaces. It was a way to manage complexity, protect invariants, and enable modularity (we going to cover the origin of encapsulation in more details later in this post).</p><p>This has always been — in my opinion — the cornerstone of good experience design, <strong>which is to remove the burden of complexity from users</strong>:</p><blockquote>Who should bear the burden of complexity within an application or a process — the user, or the designers and developers? This is a fundamental question when considering the design of user interfaces and, more broadly, how humans interact with technology. <em>Yablonski, J., “Tesler’s Law” in </em><a href="https://lawsofux.com/articles/2024/teslers-law/"><em>Laws of UX</em></a><em> (2024)</em></blockquote><p>In Human–Agent Centered Design (H-ACD), encapsulation takes on an even broader role: it becomes the mechanism by which <strong>meaning, responsibility, and agency are bundled together</strong>.</p><p>The core premise of H-ACD remains unchanged: humans are still the recipients of value. What changes is the path between human intention and system execution. Agents become intermediaries — interpreting intent, orchestrating tools, and acting at a scale and speed no human workflow could sustain.</p><p>Encapsulation is what makes this mediation possible. It collapses procedural paths, absorbs <a href="https://www.designative.info/tag/uncertainty-and-ambiguity/">ambiguity</a>, and translates intent into coordinated action. Encapsulation doesn’t just hide complexity — it reshapes what humans must understand to remain in control.</p><h3>A Short History of Design Paradigm Shifts</h3><p>Encapsulation fits into a longer arc of interaction design evolution:</p><ul><li>Task-centric HCI optimized human performance within rigid systems.</li><li>User-centered design improved usability and learnability.</li><li>Human-centered and values-based design expanded the lens to ethics, inclusion, and meaning.</li><li>Agent-mediated systems introduce autonomy as a first-class design variable.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/580/0*DasZhdrFkWew00qc" /></figure><p>This shift creates a <strong>triadic relationship</strong>:</p><p><strong>Human &lt;-&gt; Agent &lt;-&gt; System</strong></p><p>Interaction is no longer dominated by direct manipulation. Instead, it becomes an act of intent expression followed by delegated execution. The user does not operate the system step by step; they negotiate outcomes with an intermediary that acts on their behalf.</p><h3>Why Task Analysis and User Flows Still Matter — Just Not as the Primary Design Object</h3><p>It is important to be explicit about what this argument is not.</p><p>Questioning user <a href="https://www.designative.info/2025/11/20/flows-age-agentic-ai-what-if-our-core-ux-models-no-longer-apply/">flows in the age of Agentic AI</a> is not a call to abandon user research, task analysis, or Jobs-to-be-Done. On the contrary, these practices become more important — but their role changes.</p><p>In Human–Agent Centered Design, task analysis no longer exists primarily to choreograph interaction steps. Instead, it helps designers surface implicit knowledge about human intentions.</p><p>When we analyze tasks today, we are not just documenting behavior. We are uncovering:</p><ul><li>what people are trying to accomplish,</li><li>where uncertainty emerges,</li><li>what conditions trigger delegation,</li><li>and which moments require reassurance, judgment, or accountability.</li></ul><p>This is where <a href="https://www.designative.info/2025/09/24/jobs-to-be-done-and-intention-mapping-translating-human-needs-into-agent-actions/">Jobs-to-be-Done and Intention Mapping</a> become critical.</p><p>JTBD provides stable anchors — human goals that persist regardless of interface, modality, or automation level. Intention Mapping extends this logic by making the translation between those goals and autonomous system action explicit. In agent-mediated experiences, interactions become conversational negotiations rather than linear command–response sequences.</p><p>From this perspective:</p><ul><li><strong>Task analysis informs what should be encapsulated.</strong> Repeated, high-friction, or cognitively expensive task structures signal candidates for workflow encapsulation.</li><li><strong>User flows inform risk, trust, and intervention points.</strong> Even when steps disappear from the surface, flows remain essential for identifying where transparency is required, where failure is costly, and where humans must be able to interrupt or override agent behavior.</li><li><strong>Task analysis feeds intention modeling.</strong> When workflows become fluid or fully encapsulated, intention becomes the stable design unit.</li></ul><p>When tasks become fluid or entirely encapsulated, intention becomes the stable design unit. Designing for agentic systems is no longer about choreographing the journey from Screen A to Screen B. It’s about constructing a precise, resilient relationship between <strong>intent -&gt; interpretation -&gt; action -&gt; accountability</strong>.</p><p>Workflow Encapsulation removes the cognitive scaffolding that task analysis once relied on. Design must replace it with new forms of intelligibility — ones suited to delegation rather than operation.</p><p>User flows do not disappear; they recede into the background, informing the design of agent boundaries and guardrails rather than defining the experience itself.</p><h3>Guardrails, boundaries, and “when wrong” behavior</h3><p>Amershi et al.’s <a href="https://www.microsoft.com/en-us/research/wp-content/uploads/2019/01/Guidelines-for-Human-AI-Interaction-camera-ready.pdf">guidelines for human–AI interaction</a> give a concrete playbook for this shift in literacy. They show that AI systems must not only “make clear what they can do” and “what they can’t do,” but also support efficient correction, provide meaningful feedback when they are wrong, and allow users to easily reverse or refine automated actions. These empirically validated patterns map directly onto workflow encapsulation: every encapsulated flow should declare its scope, its guardrails, its failure modes, and the user’s recourse when outcomes look wrong.</p><p>In other words, once a multi-step process collapses into a single agentic act, the primary design surface is no longer a series of screens but a bundle of guarantees: <strong>what inputs are acceptable, what outputs are promised, what side effects are allowed, and how a human can intervene</strong>. Designing these guarantees is not a “nice-to-have”; it is how teams prevent encapsulation from becoming an uninspectable black box.</p><h3>What Users Now Need to Know (and What They No Longer Do)</h3><p>Encapsulation fundamentally shifts knowledge requirements.</p><p>Users no longer need procedural mastery or screen-level predictability. They no longer have to understand every step to perform work effectively.</p><blockquote>How comfortable are users about the computer knowing how they think, and how tolerant they will be if (and when) the computer makes a mistake guessing their desires? <em>Maeda, J., </em><a href="https://amzn.to/3YxmHi6"><em>The laws of simplicity</em></a><em> (2020)</em></blockquote><p>What they <em>do</em> need is different:</p><ul><li><strong>The <em>boundaries</em> of the agent’s authority</strong>: what decisions it may take autonomously, and where it must defer.?</li><li><strong>The <em>guardrails</em> that constrain its behavior</strong>: fairness, safety, or domain rules it is obligated to respect. (Hoffman, R. R., Mueller, S. T., Klein, G., &amp; Litman, J., <a href="https://arxiv.org/pdf/1812.04608">Metrics for explainable AI</a>, 2018)</li><li><strong>The <em>fallbacks</em> that apply under uncertainty</strong>: when it will ask, when it will abstain, and how it recovers from error (Bussone, A., Stumpf, S., &amp; O’Sullivan, D., <a href="https://dl.acm.org/doi/10.1109/ICHI.2015.26">The role of explanations on trust and reliance in clinical decision support systems</a>, 2015).</li><li><strong>The <em>escalation</em> and <em>override paths</em></strong> that govern breakdowns: when responsibility shifts back to humans, how exceptions are handled, and how trust is repaired after failure (Noessel, C., <a href="https://amzn.to/3JWyoMk">Designing Agentive Technology</a>, 2017).</li></ul><h3>Workflow Encapsulation in Human-Agent Centered Design</h3><p>Encapsulation in Human–Agent Centered Design is not a new concept; it is a deliberate reuse of a term with a long, precise history in computer science and artificial intelligence. What changes in H-ACD is the object of encapsulation: <strong>instead of just hiding code and data structures, designers are encapsulating meaning, responsibility, and control in the way agents act on behalf of humans</strong> (Horvitz, E., <a href="https://dl.acm.org/doi/10.1145/302979.303030">Principles of mixed-initiative user interfaces</a>, 1999).</p><h3>Where “encapsulation” comes from</h3><p>In software engineering, encapsulation originally described the bundling of data and operations into modules that expose a clear interface while hiding internal details, so those modules can change without breaking the wider system. Classic accounts emphasize its role in managing complexity and enabling the independent evolution of components, particularly in large, collaborative software projects (IEEE Standards Coordinating Committee, <a href="https://ieeexplore.ieee.org/iel1/2238/4148/00159342.pdf">IEEE standard glossary of software engineering terminology</a>, 1990)</p><p>In AI and machine learning, related ideas include <em>abstraction</em> and <em>hierarchical control</em>: agents operate on higher-level “chunks” of behavior rather than raw, low-level actions. Work on temporal abstraction and options in reinforcement learning demonstrates how <strong>multi-step action sequences can be compressed into macro-actions, which are treated as single decisions at a higher level</strong> (Sutton, R. S., Precup, D., &amp; Singh, S., <a href="http://incompleteideas.net/papers/SPS-98.pdf">Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning</a>, 1999).</p><h3>From code abstraction to workflow abstraction</h3><p>This lineage is exactly what H-ACD borrows. When designers operating the Human-Agent Centered Design framework talk about “encapsulating a workflow,” the analogy is to those macro-actions and modules: a <strong>messy, multi-step process is bundled into a single callable capability that an agent can invoke and coordinate internally</strong>. The user no longer sees or configures every intermediate step; what they encounter is a higher-level “unit of work” with a clear purpose and a bounded surface.</p><p>Systems are valued for reducing cognitive and technical burdens. Encapsulated tools enable agents to execute processes end-to-end, while humans retain the authority to validate, supervise, and decide. This expresses a value of shared responsibility between automation and human oversight.</p><p>HACD, <a href="https://hacd.lovable.app/">Human-Agent Centered Design</a> (2025)</p><p>However, the design focus is different. In engineering, success is often measured by modularity, reuse, and ease of maintenance. In H-ACD, success is measured by whether humans can still understand what the agent is doing on their behalf, intervene when needed, and hold someone accountable when things go wrong.</p><p><strong>Workflow encapsulation compresses multi-step processes into a single agent-triggered action</strong>.</p><p>Consider an enterprise HR scenario:</p><blockquote><em>“Generate and schedule this month’s compliance reminders.”</em></blockquote><p>Behind this simple request sits a complex orchestration:</p><ul><li>eligibility checks</li><li>content generation</li><li>approval routing</li><li>scheduling</li><li>logging and auditing</li></ul><p>From the user’s perspective, the flow disappears. Sequencing is no longer explicitly controlled. Steps may never be shown.</p><blockquote><em>The flow disappears.<br>The intention remains.<br>Meaning must remain.</em></blockquote><p>Chris Noessel describes trust in agentive systems as a “looping, careening rollercoaster”: <strong>slow to build, fragile to maintain, and quick to collapse</strong>.</p><p>Workflow encapsulation intensifies this dynamic. <strong>When many steps are compressed into a single delegated action, users are no longer granting trust to individual interactions, but to the entire orchestration behind the scenes</strong>. Each successful outcome may earn incremental trust, but a single opaque failure, unexplained decision, or poorly handled exception can undo it all at once. This is why, in encapsulated workflows, exception handling, escalation paths, and recovery mechanisms are not edge cases — they are the primary surfaces through which trust is won, lost, and rebuilt over time.</p><p>Wilson and Daugherty’s notion of augmentation (2018) helps explain <em>why</em> Noessel’s “looping, careening rollercoaster of trust” is such a persistent pattern. <strong>As AI systems move from simple amplification to interaction and, eventually, embodiment, they continuously redistribute agency between humans and machines — reshaping not just interfaces, but how work itself gets done</strong>. We will cover Wilson and Daugherty’s augmentation concept in details later in this article.</p><p>Noessel reminds us that trust spikes when this redistribution feels helpful and collapses when it becomes opaque or surprising. Seen together, the problem is not merely one of UX polish or communication; trust instability is structural. It emerges when systems change what humans are responsible for without clearly signaling how control, accountability, and judgment have shifted — making guardrails at the tool and agent level essential, not optional.</p><h3>Workflow Encapsulation as a Governance Construct</h3><p>This is where H-ACD extends the original notion: encapsulation becomes not merely a means of hiding complexity but a means of governing it. When a workflow is encapsulated, the interface is no longer just a function signature; it is a contract that specifies (Shneiderman, B., <a href="https://amzn.to/3KuBOX2">Human-centered AI</a>, 2022):</p><ul><li><strong>What human intention does this unit serve</strong>?</li><li><strong>What authority does the agent have while executing it</strong>?</li><li><strong>What guarantees and guardrails are in place</strong>?</li><li><strong>How can humans observe, override, or audit the outcome</strong>?</li></ul><h3>Building Trust Inside Encapsulated Workflows</h3><p>When workflows are encapsulated, users no longer see every step an agent takes, so trust must be established by the <em>contract</em> around that encapsulation: what the agent is allowed to do, how it behaves when uncertain, and how humans can stay in charge. For designers, this means trust <strong>can’t be an afterthought layered on top of an agentic feature; it has to be designed into the boundaries, explanations, and recovery paths of each encapsulated tool</strong>.</p><p>As Chris Noessel observed in <em>Designing Agentive Technology</em>:</p><p>Trust in agents is built slowly across many successful interactions and can collapse quickly with a single failure — making exception handling and escalation not edge cases, but central design concerns.</p><p>Noessel, C., <a href="https://amzn.to/3JWyoMk">Designing Agentive technology</a> (2022)</p><p>In practice, there are three conversations designers should drive with PMs and engineers during workflow encapsulation:</p><p><strong>Authority and visibility</strong></p><ul><li>What is the agent <em>explicitly</em> allowed to decide autonomously, and what must always be confirmed by a human?</li><li>For each encapsulated workflow, what must the user be able to <em>see</em> and <em>understand</em> to feel in control (before, during, and after the action)? Designers should capture this as clear authority boundaries, default levels of autonomy, and mandatory “glanceable” status indicators.</li></ul><p><strong>Uncertainty, error, and recovery</strong></p><ul><li>How will the agent signal low confidence or ambiguous situations instead of silently guessing?</li><li>What are the expected fallback behaviors (ask for clarification, reduce scope, defer, escalate), and how obvious are they to the user in the moment? These answers belong directly in the encapsulated tool model: confidence thresholds, escalation rules, and designed recovery paths that preserve dignity and minimize harm.</li></ul><p><strong>Explanations and evaluation of trust over time</strong></p><ul><li>For each major action inside the workflow, when is an explanation required, and for whom (end users, support, audit)?</li><li>Which trust-related metrics will be monitored (e.g., perceived control, appropriate reliance, override rates, recovery success), and what thresholds trigger a design or policy review? Designers should ensure that explanation requirements and <a href="https://www.designative.info/2025/12/01/dual-evaluation-measuring-both-human-experience-and-agent-effectiveness/">Dual Evaluation</a> metrics are recorded alongside functional specs, so trust becomes measurable and debatable — not just a brand promise.</li></ul><p>If designers consistently surface these questions in encapsulation work, trust stops being a vague property of “the AI” and becomes a set of explicit, testable commitments embedded in each workflow: what the agent will do, how it will show its work, and how humans can always reclaim control when it matters.</p><h3>Which Workflows Deserve Encapsulation?</h3><p>Designers are often told to “embrace automation,” but rarely invited into the conversation about <em>where</em> automation actually belongs. <strong>If encapsulation lets agents pull entire workflows “inside the box,” the strategic question is no longer just <em>how</em> to design these agents — it is <em>which</em> workflows should be encapsulated in the first place</strong>. Some sequences are ripe for delegation; others are fundamentally about human judgment, legitimacy, or relationship and break when collapsed into a single opaque action.</p><p>This is where designers need to be empowered, not sidelined. The role is not to rubber-stamp automation decisions, but to ask the uncomfortable, high-leverage questions: Is this really a machine’s job? What value lives in the human process itself? What risks do we create if this flow disappears from view? <strong>Framed that way, “Are there workflows that are more suitable to be encapsulated than others?” is not a technical detail</strong> — it’s a product strategy question that design is uniquely positioned to lead.</p><p>Wilson and Daugherty’s “missing middle” concept (2018) is a useful counterpoint to the simplistic story that AI will either replace humans or leave their work untouched. <strong>It describes a broad band of hybrid activities where humans and machines complement each other</strong>: people lead, empathize, and set direction, while AI handles pattern recognition, iteration, and optimization at scale.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/580/0*277awn5gJQ95owRE" /><figcaption>In the missing middle, humans work with smart machines to leverage each party&#39;s strengths. Humans, for example, are needed to develop, train, and manage various Al applications. In doing so, they are enabling those systems to function as true collaborative partners. For their part, machines in the missing middle are helping people to punch above their weight, providing them with superhuman capabilities, such as the ability to process and analyze copious amounts of data from myriad sources in real time. Machines are augmenting human capabilities (Wilson, H. J., &amp; Daugherty, P. R., <a href="https://amzn.to/45lM60b">Human + machine: Reimagining work in the age of AI</a>, 2018)</figcaption></figure><p>Seen through the lens of the missing middle, augmentation is not a single mode of collaboration but a spectrum of roles that agents play within hybrid workflows. The design challenge is not whether AI is present, but how its agency is scoped, sequenced, and governed so that human and machine strengths remain complementary:</p><ul><li><strong>Amplification</strong>: Agents extend human cognition without replacing human judgment. They analyze data, explore alternatives, and surface patterns at scale, while authority and accountability remain firmly human. This is the “punch above your weight” zone of the missing middle.</li><li><strong>Interaction</strong>: Agents act on behalf of humans or mediate exchanges between people and systems. They interpret intent, manage dialogue, and operate at scale. Here, trust depends on clear boundaries, graceful recovery, and well-defined escalation when uncertainty or risk increases.</li><li><strong>Embodiment</strong>: Agents operate in physical, financial, or operational reality — not just interfaces. They take action, not merely make recommendations. This is the most consequential part of the missing middle, demanding tight coupling between delegated authority, oversight, and feedback to prevent errors from propagating into the real world.</li></ul><p>Together, these modes show why the missing middle is an orchestration problem rather than a substitution problem: the goal is not full automation, but deliberate, governed collaboration.</p><h3>Workflows prime for encapsulation</h3><p>These are usually <strong>machine-strong, human-brittle</strong> activities where people add little value by micromanaging steps:</p><ul><li><strong>High-volume, rules-heavy sequences</strong> like eligibility checks, routing, formatting, scheduling (Bhavsar, K., <a href="https://www.aubergine.co/insights/designing-for-ai-agents">Designing for AI agents</a>, 2025)</li><li><strong>Stable processes with clear success criteria</strong> and rich historical data for learning (Ruh AI., <a href="https://www.ruh.ai/blogs/hierarchical-agent-systems">Hierarchical AI agent systems</a>, 2025).</li><li><strong>Tasks where latency matters more than human nuance</strong> like fraud screening passes, inventory checks, retry logic (Google Cloud, <a href="https://cloud.google.com/architecture/choose-design-pattern-agentic-ai-system">Reason and act (ReAct) pattern</a>, 2025).</li></ul><p>For these, encapsulation turns multi-step flows into callable capabilities: <strong>the agent handles sequencing, retries, and optimization inside the box, while humans define boundaries, guardrails, and overrides at the edges</strong>.</p><h3>Workflows that should stay hybrid</h3><p>The <strong>“missing middle”</strong> — where humans and machines complement each other — is the real sweet spot for workflow encapsulation.?</p><ul><li>Activities where humans <strong>lead, empathize, judge, and create direction</strong>, and agents support with analysis, generation, or orchestration like campaign design, diagnosis support, scenario planning (Bhavsar, K., <a href="https://www.aubergine.co/insights/designing-for-ai-agents">Designing for AI agents</a>, 2025).</li><li><strong>Flows with meaningful ambiguity or value trade-offs</strong>, where human judgment and accountability must remain foreground (Starke, A. I., Van der Linden, S., &amp; colleagues, <a href="https://doi.org/10.1016/j.patter.2024.xxx">Establishing and evaluating trustworthy AI: Overview and research challenges</a>, 2024).</li></ul><p>Here, encapsulation should focus on <strong>subflows</strong>, not the whole journey: research synthesis, candidate generation, risk scoring, simulation. <strong>The human remains the decision-maker; the agent encapsulates the heavy lifting beneath that decision</strong>.</p><h3>Workflows that are poor candidates for deep encapsulation</h3><p>Some activities are <strong>human-only by design</strong> and should resist heavy encapsulation:</p><ul><li><strong>Where the primary value is relational or ethical</strong>, like coaching, performance reviews, sensitive negotiations (European Commission High-Level Expert Group on Artificial Intelligence, <a href="https://digital-strategy.ec.europa.eu/en/library/ethics-guidelines-trustworthy-ai">Ethics guidelines for trustworthy AI. Publications Office of the European Union</a>, 2019).</li><li><strong>Where legitimacy depends on visible, human-led process</strong>, like governance boards, disciplinary decisions (Mittelstadt, B. D., <a href="https://doi.org/10.1038/s42256-019-0114-4">Principles alone cannot guarantee ethical AI</a>, 2019)</li></ul><p>Automation can still support (e.g., drafting, retrieval), but the workflow itself should not be collapsed into a single opaque agent action. In these cases, design for <strong>augmentation</strong>, not encapsulation.</p><h3>A practical rule of thumb for designers</h3><p>When deciding whether to encapsulate a workflow, ask:</p><ol><li><strong>Is the value in the outcome, or in the human <em>process</em>?</strong> If process visibility is part of trust or legitimacy, be cautious</li><li><strong>Is human judgment central, or is it mostly coordination and grunt work?</strong> Encapsulate the grunt work; keep judgment in the open.</li><li><strong>Can we define clear success criteria, guardrails, and recovery behaviors?</strong> If not, the workflow is not ready to be safely encapsulated.</li></ol><p>Encapsulation works best where agents can reliably “do the doing” inside the box, and humans still own the intent, the boundaries, and the meaning of the outcome outside it.</p><p><em>This post is adapted from a longer piece in an ongoing series on Human-Agent Centered Design, </em><a href="https://www.designative.info/2025/12/16/workflow-encapsulation-in-human-agent-centered-design-from-doing-work-to-governing-outcomes/"><em>published in full on my blog</em></a><em>.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*n21jBb7PW-K-592L" /><figcaption>From flows to intent surfaces: H-ACD’s <a href="https://www.designative.info/2025/12/16/workflow-encapsulation-in-human-agent-centered-design-from-doing-work-to-governing-outcomes/">Encapsulated Workflow</a> Principle defines how Agents can execute on work end-to-end, while human authority remains central to oversight and decision-making.</figcaption></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=de416b27226d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-bootcamp/workflow-encapsulation-in-human-agent-centered-design-from-doing-work-to-governing-outcomes-de416b27226d">Workflow Encapsulation in Human–Agent Centered Design: From Doing Work to Governing Outcomes</a> was originally published in <a href="https://medium.com/design-bootcamp">Bootcamp</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Rethinking Design Critiques in the Age of AI Prototyping]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://medium.com/design-bootcamp/rethinking-design-critiques-in-the-age-of-ai-prototyping-9e8c5bab5b0b?source=rss-6c2b386e05b2------2"><img src="https://cdn-images-1.medium.com/max/1200/0*7sYWmXHx4wkZqgws" width="1200"></a></p><p class="medium-feed-snippet">How design critiques must evolve to balance the speed of AI-driven prototyping with vision, intent, and shared understanding.</p><p class="medium-feed-link"><a href="https://medium.com/design-bootcamp/rethinking-design-critiques-in-the-age-of-ai-prototyping-9e8c5bab5b0b?source=rss-6c2b386e05b2------2">Continue reading on Bootcamp »</a></p></div>]]></description>
            <link>https://medium.com/design-bootcamp/rethinking-design-critiques-in-the-age-of-ai-prototyping-9e8c5bab5b0b?source=rss-6c2b386e05b2------2</link>
            <guid isPermaLink="false">https://medium.com/p/9e8c5bab5b0b</guid>
            <category><![CDATA[design-critique]]></category>
            <category><![CDATA[prototyping]]></category>
            <category><![CDATA[ai-ux-design]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[ux]]></category>
            <dc:creator><![CDATA[Itamar Medeiros]]></dc:creator>
            <pubDate>Fri, 19 Dec 2025 12:11:26 GMT</pubDate>
            <atom:updated>2025-12-19T12:11:26.445Z</atom:updated>
        </item>
    </channel>
</rss>