<?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 Ana Siro on Medium]]></title>
        <description><![CDATA[Stories by Ana Siro on Medium]]></description>
        <link>https://medium.com/@ana.siro?source=rss-0fef1576603b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*WcKQ38m7Os-cfzXy</url>
            <title>Stories by Ana Siro on Medium</title>
            <link>https://medium.com/@ana.siro?source=rss-0fef1576603b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 02:34:14 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ana.siro/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[Designing the Invisible: The Role of Service Design in Agentic Systems.]]></title>
            <link>https://medium.com/design-globant/designing-the-invisible-the-role-of-service-design-in-agentic-systems-3b7e71a9d44d?source=rss-0fef1576603b------2</link>
            <guid isPermaLink="false">https://medium.com/p/3b7e71a9d44d</guid>
            <category><![CDATA[agentic-design]]></category>
            <category><![CDATA[ai-design]]></category>
            <category><![CDATA[globant]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[service-design]]></category>
            <dc:creator><![CDATA[Ana Siro]]></dc:creator>
            <pubDate>Fri, 10 Apr 2026 19:31:58 GMT</pubDate>
            <atom:updated>2026-04-10T19:31:58.555Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="A vibrant conceptual illustration on a bright green background. A multi-armed figure with a single giant eye for a head acts as a conductor. They are orchestrating floating, chaotic abstract shapes and an hourglass on their left, and transforming them into structured documents, graphs, and checklists on their right, ultimately leading to a glowing blue star. A small black cat sits at their feet." src="https://cdn-images-1.medium.com/max/1024/1*tzRLCetl90dwB2CV1kYZHw.png" /><figcaption>Illustration by Luciana Carlassara</figcaption></figure><p><em>We all knew AI was coming, but suddenly, it was everywhere. The focus evolved: The question is no longer simply, “How can I use ChatGPT or Gemini to aid my design process?” but rather, “</em><strong><em>How do I orchestrate a system of autonomous agents to deliver value that is real and scalable?</em></strong><em>”. This presented us as designers with two major challenges: mastering the use of AI tools and developing the ability to design complete AI solutions.</em></p><p>At Globant, we faced this challenge at scale: a community of 2,000 professionals, including 700 designers and 1,300 product specialists, each operating within different AI contexts across projects. To standardize design practices in this new era, we turned to our core strength: Strategic Design.</p><h3>I am a Service Designer at Globant, and <strong>this is a look into how, </strong>together with an incredible team, we built an agentic ecosystem in less than six months.</h3><p>Our goal was simple: <strong>helping our teams get better results in less time, letting them focus on what really matters</strong>. To get there, we navigated <strong>a process that prioritized</strong> <strong>listening to users at every stage</strong> through <strong>workshops, surveys, and testing</strong>.</p><blockquote><strong>I firmly believe that design, especially when building AI, must remain user-focused to ensure adoption, trust, quality, and results that actually work.</strong></blockquote><h3>1. The Map: Finding Common Ground in a World of Hundreds of Clients</h3><p>The first step was to define a single, unified Design and Product Life Cycle. Understanding that every client and project at Globant is unique, we needed a common ground to <strong>build agnostic tools that could work anywhere: any industry, any client, any product.</strong></p><p>By co-creating with managers from both Design and Product Studios, we mapped the “As-Is” process, summarizing the end-to-end flow of Discovery, Delivery, and Development. This map allowed us to understand exactly where we stood and where AI could have a systemic impact, <strong>transforming the whole workflow, not just small tasks</strong>.</p><h3>2. Discovery: Opportunity Workshops</h3><p>With the map on the table, I facilitated different <strong>ideation workshops</strong> with representatives from both Design and Product teams. The goal wasn’t “AI for the sake of AI,” but rather to attack the friction points:</p><ul><li><strong>Time-consuming tasks:</strong> Repetitive processes with low cognitive value.</li><li><strong>Bottlenecks:</strong> Points where the workflow stalls.</li><li><strong>Blue-sky opportunities:</strong> Areas where AI could completely transform the process.</li></ul><h3>3. Reality Check: Shadow AI</h3><p>Parallel to the ideation phase, we launched a <strong>massive survey across both Studios</strong> (2,000+ people). We realized everyone was already using AI. It was <strong>‘Shadow AI’</strong>, tools used in the shadows without a clear guide. Various teams were independently benchmarking AI tools, choosing those best suited for their workflows, and developing custom agents within our <strong>proprietary AI platform Gloab.ai OS and any other AI-Clients available in the market.</strong></p><p>This wasn’t just a problem of AI tools; it was a cry for a better, standardized AI process. Security was never the issue; fragmentation was.</p><h3>4. Strategy: High-Level System Definition</h3><p>Based on the data, we reached a strategic conclusion: to standardize quality outputs, we needed to create our own <strong>Agentic Systems</strong> — agnostic of specific clients or industries.</p><p>Using our <em>Life Cycle</em> map, we defined high-level systems, assigning each a clear purpose and location within the process:</p><ul><li><strong>Design Discovery Agentic System</strong></li><li><strong>Design Research Agentic System</strong></li><li><strong>Strategic Consulting Agentic System</strong></li><li><strong>Product Backlog Definition System</strong></li></ul><p><em>We defined several systems, but for this article, I want to focus on the ones we prioritized for our initial rollout.</em></p><h3>5. Multidisciplinary Tracks: Co-Creation in Sync</h3><p>We moved away from a centralized approach to organized and specialized work tracks. We formed <strong>Definition Cells</strong> — basically small, focused teams — to tackle each system. No silos, no endless meetings, just designers and engineers talking to each other, <strong>seeking the best solution for each AI need.</strong></p><p>We had one track for each Agentic System form by: Experts in that Area (Design, Product backlog, Research) and AI Specialists (Data Scientists, Data Engineers or Business Intelligence).</p><p>As the Service Designer, <strong>I acted as the “connective tissue”</strong> across all tracks. My role was to ensure that while each team focused on its specific domain, the systems maintained a unified logic. This oversight was vital to prevent agent capabilities from overlapping and to avoid rework.</p><p>This collaboration between Domain Experts and Technical Teams allowed us to define the precise anatomy of every agent:</p><ul><li><strong>Inputs</strong>: We strictly differentiated between <strong>mandatory data</strong> (required to run) and <strong>optional data </strong>(context).</li><li><strong>Outputs</strong>: Detailed definition of response format, tone, and structure.</li><li><strong>Capabilities &amp; Tools: </strong>The specific “powers” each agent needed to execute its task.</li><li><strong>Guardrails:</strong> Establishing ethical limits and operational boundaries.</li><li><strong>User Flows: </strong>Mapping the step-by-step diagram of the human-agent interaction.</li></ul><h3>6. Construction: From Definition to PoC</h3><p>The “handoff” was a list of requirements for each agent from Domain Experts to the Technical Teams, which was reviewed and built together. With the specifications approved, we moved to the<strong> Proof of Concept (PoC) construction phase.</strong></p><p>The construction was a <strong>side-by-side engineering process</strong>. While engineers assembled the agents in our proprietary <strong>Gloab.ai OS</strong> and Gemini <strong>Gems</strong>, the designers validated the logic in real-time. If an agent wasn’t retrieving or generating the correct information, we adjusted the definition or the system prompt on the spot. This technical iteration was vital to transform theoretical definitions into functional AI tools.</p><h3>7. Validation: UAT (User Acceptance Testing)</h3><p>Once the PoCs were built, we moved to rigorous testing:</p><ul><li><strong>Unit Testing:</strong> We tested agent-by-agent to validate individual precision and the quality of the outputs.</li><li><strong>System Testing:</strong> We validated the entire flow to ensure agents could effectively “talk” to each other, ensuring the output of one agent served as a valid input for the next.</li></ul><h3>8. Pilot: Friends &amp; Family</h3><p>Before the mass release, we ran a “fire test”: a <strong>3-week pilot with 3 real design project teams</strong> using the <strong><em>Design Discovery Agentic System</em></strong>. This allowed us to measure real-world adoption, the learning curve, and the perceived utility in live projects, ultimately validating the readiness of the agents before launching to the entire Design Studio.</p><p>We did the same with the <strong>Product Backlog Definition System, </strong>with 11 Product Managers and Business Analysts testing the agents in their projects.</p><h3>9. Enablement &amp; Release</h3><p>Finally, we prepared for launch. We created a <strong>Playbook</strong> that went beyond technical instructions; it taught teams how to integrate AI into their cognitive workflow.</p><h4><strong>A Collective Milestone</strong></h4><p>This ecosystem was a collective endeavor. While the core team was small, we achieved this massive rollout in less than six months thanks to the support of countless colleagues.</p><h3>This speed was only possible because of Globant’s culture of collaboration.</h3><p>This system wasn’t just built for the Designers and Product teams; it was built by them.</p><blockquote>This journey confirmed that designing Agentic Systems is not a purely technical task; it is also a <strong>design practice</strong>, as it requires a<strong> deep understanding of human processes</strong>. By uniting Domain Experts with Data Engineers, we created true digital teammates that enhance our creative capacity.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3b7e71a9d44d" width="1" height="1" alt=""><hr><p><a href="https://medium.com/design-globant/designing-the-invisible-the-role-of-service-design-in-agentic-systems-3b7e71a9d44d">Designing the Invisible: The Role of Service Design in Agentic Systems.</a> was originally published in <a href="https://medium.com/design-globant">Design Globant</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>