<?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 Joerg Beringer on Medium]]></title>
        <description><![CDATA[Stories by Joerg Beringer on Medium]]></description>
        <link>https://medium.com/@joerg.beringer?source=rss-ba38a7cfefbf------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*E51iRduFVHyk5un7gcmIog.png</url>
            <title>Stories by Joerg Beringer on Medium</title>
            <link>https://medium.com/@joerg.beringer?source=rss-ba38a7cfefbf------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 22:29:01 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@joerg.beringer/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[Self-Referential Research]]></title>
            <link>https://medium.com/@joerg.beringer/self-referential-research-088d0a36b7df?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/088d0a36b7df</guid>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[writing-prompts]]></category>
            <category><![CDATA[user-research]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Tue, 28 Jul 2026 09:41:36 GMT</pubDate>
            <atom:updated>2026-07-28T09:41:36.522Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-fyP3OAZyxT8jgmP3UvL3w.png" /></figure><h4>Self-referential design often begins with self-referential research.</h4><h3>Introduction</h3><p>Back in the 90s and early 2000s, Alan Cooper coined the term “self-referential design”. He was referring to the risk of designers designing for themselves rather than for their audience. If there is no external frame of reference, designers will simply design the solution the way they like it or are familiar with. While Alan Cooper was referring to the individual designer, the same phenomenon exists at multiple levels: at the organizational level when defending an existing product, at the workshop level when selecting participants, in user research when asking leading questions, and today, increasingly, in AI-assisted user research when writing leading prompts.</p><h3>Insight-out thinking in product management</h3><p>Domain experts in organizations slowly become disconnected from the actual work practice, product owners see their existing product as the solution to every problem, and existing products are incrementally improved within the established solution space instead of studying the actual user needs. This is also part of the innovator’s dilemma because successful product organizations become immune to discovering new solutions, we at ProContext call this the “immunization trap” preventing innovation.</p><blockquote>The moment requirements are derived from existing solutions instead of observed user needs, user research becomes self-referential.</blockquote><h3>Design Thinking helps — but can turn into inside-in thinking</h3><p>One way to avoid this trap is to get closer to the end users and start observing the actual users without imposing a solution on them. The Gemba Walk or Contextual Design are two famous methods of this kind of ethnographic field research. However, because such field research is expensive, time-consuming, and requires skilled user researchers, it is often replaced by internal workshops that use design thinking exercises to immerse themselves in the user’s point of view and generate new ideas by leveraging the existing knowledge of internal stakeholders and invited customer representatives who act as proxy users. The inherent risk is that these participants may also be immunized based on their affiliation with the product organization that invites them to the design thinking workshop.</p><h3>Can a handful of stakeholders provide representative input?</h3><p>One of the main goals of these workshops is to break away from the day-to-day business and rethink the solution in a fresh and unbiased way. While these workshops seed innovative ideas, they are statistically dependent on a small set of stakeholders who are often subject to the collective immunization trap of innovation described above. Therefore, workshops need to be grounded in real customer/user data, which is often done by having domain experts share their knowledge about a particular work practice as input to the workshop.</p><p>The quality of the input depends on how representative this input is and how much these “proxy users” are biased by their own organization and experience. Replacing the designer’s tendency to self-referentially design with the proxy user’s tendency to self-referentially remember is still a high risk. This is why design thinking introduces a lot of forced voting and shortlisting, so that at least all workshop participants have the same decision-making power.</p><blockquote><strong>The moment participant selection reflects organizational assumptions rather than the context of use, user research becomes self-referential.</strong></blockquote><p>This dependency on participants is an inherent risk of design thinking workshops. While 3–5 field interviews result in a solid, unbiased picture of the user’s reality, 3–5 workshop participants do not guarantee an unbiased view of the world. At the same time, locking these participants in a room for 2–4 days is also quite expensive in terms of man-hours from an organizational perspective.</p><h3>Leading prompts introduce a new form of bias</h3><p>The growing adoption of AI in user research promises to dramatically reduce the effort required to synthesize contextual information. However, it also introduces a new source of bias that is often overlooked: <strong>leading prompts</strong>.</p><p>Just as leading interview questions can unintentionally steer participants toward a particular answer, prompts can steer a language model toward a particular interpretation of a product context. Every prompt carries implicit assumptions about what is important, which entities should be considered, and which relationships deserve attention. Two user researchers using the same language model can therefore arrive at different descriptions of the same work practice not because the context differs, but because their prompts lead the analysis in different directions.</p><blockquote><strong>The moment AI is prompted from a self-referential perspective, it will amplify that perspective.</strong></blockquote><p>Without a consistent methodology, AI-assisted user research risks replacing interviewer bias with prompt bias. Instead of asking <em>leading questions</em>, we begin writing <em>leading prompts</em>.</p><h3>Standardizing the construction of context models</h3><p>Large language models already contain extensive knowledge about common work practices across industries. The challenge is no longer accessing that knowledge. The challenge is ensuring that discovery is guided by a consistent methodology rather than by the individual prompting style of the researcher.</p><p>Product Context Analyzer (PCA) addresses this challenge by replacing ad-hoc prompting with a standardized discovery process. Instead of relying on idiosyncratic prompts to determine what should be explored, PCA applies a human-centered ontology that systematically identifies stakeholders, tasks, task objects, goals, needs, pain points, and the relationships between them. The ontology bounds the discovery space while ensuring that all relevant aspects of the context of use are explored consistently.</p><blockquote><strong>With Product Context Analyzer (PCA), user researchers remain in control of the research focus, while PCA governs the process by which the context of use is explored and transformed into an explicit reference model of the context of use.</strong></blockquote><p>This approach does not eliminate bias, nor does it replace field research or expert judgment. Instead, it replaces self-referential prompting with a governed discovery process that makes contextual understanding more systematic, transparent, and repeatable. The resulting reference model provides a common foundation from which personas, user journeys, JTBD job maps, stakeholder maps, user requirements, user stories, and other user research artifacts can be derived consistently, ensuring that every view describes the same underlying context rather than a different prompting strategy.</p><h3>Turning Bias into Managed Discovery</h3><p>Rather than replacing field research, workshops, or expert judgment, Product Context Analyzer is designed to become part of the user research process itself. By standardizing how the context of use is explored before empirical research begins, PCA establishes a common baseline that is less influenced by individual prompting styles, organizational assumptions, or local research practices.</p><blockquote><strong>The value of standardization isn’t that everyone reaches the same conclusions. It’s that everyone begins from the same methodological foundation.</strong></blockquote><p>Organizations that standardize on Product Context Analyzer create a shared methodological foundation for user research. Instead of every user researcher constructing a different model of the same product context, teams begin from the same human-centered reference model, making contextual understanding less susceptible to inside-out assumptions and allowing user researchers to focus their expertise on validating, refining, or challenging it through empirical user research.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=088d0a36b7df" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[See the Forest, Not Just the Trees]]></title>
            <link>https://medium.com/@joerg.beringer/see-the-forest-not-just-the-trees-ac69d3691b98?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/ac69d3691b98</guid>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[product-design]]></category>
            <category><![CDATA[service-design]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Tue, 21 Jul 2026 14:11:20 GMT</pubDate>
            <atom:updated>2026-07-21T14:11:20.775Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*703MfJS6cYKIfRUGPJH0KA.png" /><figcaption>Interactive Forest Map With Layers</figcaption></figure><h4>Understanding the Problem Space Through Multiple Perspectives</h4><p>We’re seeing many User Research organizations transition from rejecting AI to embracing it. Much of today’s discussion focuses on using AI to accelerate traditional research activities, such as clustering observational notes, summarizing interviews, or generating deliverables like personas, journey maps, or Jobs-to-be-Done analyses.</p><p>These tools are genuinely valuable. They help researchers produce individual deliverables faster and often with impressive quality.</p><p>But each deliverable captures only one aspect of the problem space. It’s a bit like telling stories about individual trees without ever understanding the forest they belong to.</p><p>As a result, research deliverables are often generated independently from different prompts or specialized AI tools. They may each be useful on their own, but they still need to be reconciled into a coherent understanding of the underlying context.</p><p>We at ProContext believe the process should work the other way around.</p><blockquote><strong>First map out the forest. Then capture the details of every tree. Then explore the ecosystem from any perspective.</strong></blockquote><p>That’s the idea behind our Product Context Analyzer (PCA), our AI-assisted user research tool. Instead of asking AI to generate individual research deliverables, PCA first discovers and models the relevant problem space by building a human-centred reference model of the context of use.</p><p>This is enabled by a human-centered ontology that has been developed and refined over many years. Rather than treating research deliverables as isolated documents, it captures the fundamental entities of the problem space and the relationships between them, enabling multiple research perspectives to be derived consistently from the same underlying model.</p><p>Once that foundation has been established, it generates consistent research deliverables as projections of that reference model.</p><p>These research deliverables include:</p><ul><li>Task models</li><li>User journeys</li><li>Personas</li><li>Jobs-to-be-Done analyses</li><li>Empathy maps</li><li>Stakeholder maps</li><li>User needs</li><li>User requirements</li><li>User story maps</li><li>Draft PRDs</li><li>Conceptual models</li></ul><blockquote><strong>Most UX research deliverables are not new knowledge. They are different ways of organizing the same underlying understanding.</strong></blockquote><p>Think of the human-centred reference model as a map of the forest. It captures the whole landscape once first while preserving how its different parts are connected. You can explore those relationships, inspect individual elements in detail, or step back to understand the problem space as a whole.</p><p>PCA brings this concept to life as an interconnected, hyperlinked human-centred reference model that allows UX teams to navigate relationships, inspect individual entities, and generate consistent research deliverables from the same shared foundation. Instead of spending time reconciling disconnected outputs, teams can focus on interpreting the results, refining the underlying context model, and discussing what it means for the product.</p><blockquote><strong>PCA is not just about automating the creation of research deliverables. It changes how they are created and how they fit together.</strong></blockquote><p>Product Context Analyzer is not intended to replace free-style prompting or specialized AI tools. We expect assume organizations to will continue using both. Free-style prompting remains an excellent way to explore ideas, ask questions, and investigate unfamiliar problem spaces. Purpose-built AI skills will continue to solve specialized tasks.</p><p>By rapidly mapping the problem space, PCA reduces the need to begin every project with a field study. Instead, organizations can focus their research efforts on validating assumptions, closing knowledge gaps, and investigating areas where direct observation will create the greatest value.</p><p>Once the problem space has been mapped, there is no need to collect new data every time a different research deliverable is needed. Journey maps, task analyses, user stories, requirements, and other perspectives can all be derived consistently from the same human-centered foundation.</p><blockquote><strong>Field research should validate understanding — not recreate it.</strong></blockquote><p>Likewise, not every new problem space requires starting with a field study. By rapidly mapping the problem space from knowledge already available through LLMs or interview transcripts, PCA helps organizations identify where direct observation is truly needed. Field research becomes a targeted activity for validating assumptions, closing knowledge gaps, and enriching the existing understanding rather than starting discovery from scratch.</p><p>By adopting PCA, organizations are not only adopting AI, but also establishing a repeatable, human-centered discovery practice aligned with internationally recognized ISO standards. By aligning ResearchOps with ISO 25063 (Context of Use), ISO 25064 (User Needs), and ISO 25065 (User Requirements Specification), they create a consistent foundation for discovering, documenting, and evolving knowledge across products and projects.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ac69d3691b98" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part III - Intent Lakes: The Infrastructure of AI-Native Product Engineering]]></title>
            <link>https://medium.com/@joerg.beringer/part-iii-intent-lakes-the-infrastructure-of-ai-native-product-engineering-be31433d86ee?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/be31433d86ee</guid>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[user-research]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Fri, 17 Jul 2026 13:20:13 GMT</pubDate>
            <atom:updated>2026-07-17T13:20:13.633Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YHz2n974CqXia3QlQkQ6ig.png" /></figure><h4><em>The final article in a trilogy exploring how AI is changing product engineering - not by replacing engineering, but by changing what becomes the primary engineering asset.</em></h4><h3>The Missing Architecture</h3><p>The previous article concluded with a simple observation.</p><p>If AI increasingly derives specialized solutions from explicit intent, the engineering challenge is no longer limited to generating better solutions. It becomes equally important to discover, refine, govern, curate, and evolve the intent from which those solutions are generated.</p><p>That is no longer a prompting problem. It is an architectural one.</p><p>The question is no longer how we instruct AI.</p><p>The question is how product organizations produce trustworthy intent.</p><p>For decades, software engineering has invested enormous effort in organizing implementation artifacts. Source code repositories, version control, CD/CI pipelines, design systems, API catalogs, and test automation all represent mature architectures for managing different classes of engineering assets.</p><p>Surprisingly little attention has been given to organizing the artifacts that explain why systems exist.</p><p>Stakeholder needs live in research repositories. Business objectives reside in planning tools. Regulatory constraints appear in compliance documents. Product goals are maintained elsewhere. Design rationale, customer interviews, field observations, and operational knowledge are scattered across yet more systems.</p><p>Every discipline contributes pieces of understanding.</p><p>Few organizations maintain a coherent picture.</p><blockquote><strong>Fragmented intent produces fragmented solutions.</strong></blockquote><p>That fragmentation mattered less while implementation remained the primary engineering challenge.</p><p>AI changes that equation.</p><h3>From Documentation to a Living Asset</h3><p>One implication of the previous article deserves to be taken seriously.</p><p>Intent-oriented engineering artifacts are no longer created simply to document understanding. As AI increasingly consumes explicit intent to generate specialized solutions, they become the primary interface between people and AI. That fundamentally changes their role.</p><p>Instead of producing isolated artifacts for individual projects, product organizations increasingly need a living body of explicit intent that can be reused across products, teams, disciplines, and future generations of AI systems.</p><blockquote><strong>When intent becomes the primary interface to AI, intent itself becomes a strategic engineering asset.</strong></blockquote><p>The objective is no longer merely to capture knowledge.</p><p>It is to operationalize it.</p><p>At this point, it is natural to imagine another document. An intent document. A master specification. Perhaps an AI-ready successor to today’s Product Requirements Document.</p><p>I believe that would simply repeat the pattern we are trying to leave behind.</p><p>Documents capture a moment in time. Engineering is continuous. Products evolve. User needs evolve. Business objectives evolve. Technical constraints evolve. Our understanding evolves.</p><blockquote><strong><em>I</em>ntent is produced continuously and available whenever it is needed<em>.</em></strong></blockquote><p>If intent itself continuously evolves, why should its primary representation remain a static document?</p><p>A different metaphor seems more appropriate.</p><h3>The Intent Lake</h3><p>Data engineering faced a remarkably similar challenge. Rather than forcing every source of information into a predefined structure before it became useful, organizations developed data lakes: architectures that continuously ingest information from many sources while allowing different consumers to derive different views for different purposes.</p><p>Product engineering is beginning to require something similar.</p><p>Not another documentation platform.</p><p>Not another repository of implementation artifacts.</p><p>An Intent Lake.</p><p>The Intent Lake is becoming the engineering foundation for explicit intent. It brings together stakeholder requirements, business objectives, human intent, constraints, context of use, engineering knowledge, and other forms of explicit understanding into a living body of product knowledge.</p><blockquote><strong>An Intent Lake transforms documentation into living infrastructure.</strong></blockquote><p>The important idea is not centralization.</p><p>It is architecture.</p><h3>Many Producers. Many Consumers.</h3><p>The strength of an Intent Lake lies in enabling many disciplines to contribute complementary sources of explicit intent while allowing many others to consume that shared understanding in different ways.</p><p>User Research and Product Discovery contribute evidence about people, context of use, and stakeholder intent. Product Management contributes business objectives, market opportunities, and product direction. Engineering contributes technical knowledge, architectural principles, operational experience, and implementation feedback. Compliance, operations, customer support, marketing, and many other disciplines contribute their own perspectives to the same engineering asset.</p><blockquote><strong>A single source of truth across upstream and downstream disciplines.</strong></blockquote><p>That shared intent can then be consumed in many different ways. One AI may generate workflows. Another may derive user interfaces. Others may propose architectures, create APIs, generate tests, produce documentation, or support entirely new engineering activities.</p><h3>A New Engineering Inflection Point</h3><p>The Intent Lake does more than organize engineering knowledge.</p><p>It changes where engineering connects the upstream and downstream activities of product development.</p><p>Traditionally, stakeholder understanding was progressively translated into implementation-oriented artifacts before downstream engineering could begin. Stakeholder requirements were translated into system requirements. System requirements were translated into architectures, interfaces, data models, user stories, designs, and eventually software.</p><p>Generative AI changes that relationship.</p><p>Increasingly, many downstream engineering artifacts can be derived directly from sufficiently rich intent-oriented engineering artifacts. The traditional chain of implementation specifications no longer has to serve as the mandatory bridge between understanding and implementation.</p><blockquote><strong>Shared intent replaces sequential translation of requirements.</strong></blockquote><p>The result is not the disappearance of engineering artifacts.</p><p>It is their decoupling.</p><p>Intent can evolve without requiring every downstream representation to evolve in lockstep. Instead, downstream artifacts can be regenerated whenever they are needed from the same shared understanding.</p><p>The engineering process no longer revolves around translating understanding into implementation specifications. It revolves around improving the shared source from which implementation can be generated.</p><h3>An Evolving Intent Lake Ecosystem</h3><p>The Intent Lake is beginning to emerge as the shared foundation around which AI-native product engineering naturally organizes itself.</p><p>Fundamentally, it consists of three complementary roles:</p><ul><li><strong>Upstream producers</strong> discover, validate, and contribute intent-oriented engineering artifacts that enrich the shared understanding of the product domain.</li><li><strong>Intent curators</strong> organize, connect, govern, and evolve that shared body of intent, ensuring it remains trustworthy, coherent, and reusable across products, teams, and disciplines.</li><li><strong>Downstream consumers</strong> use that shared foundation to generate specialized solution artifacts, whether they are workflows, user interfaces, architectures, software, tests, documentation, or other engineering deliverables.</li></ul><p>Around these roles, different disciplines naturally assume stewardship for different parts of the engineering process.</p><p>User Research and Product Discovery steward stakeholder understanding, desired outcomes, and context of use.</p><p>Product Management stewards the Intent Lake by connecting, prioritizing, refining, and evolving shared intent while aligning customer value, business objectives, and product direction.</p><p>Engineering stewards the realization of intent, contributes implementation knowledge and operational feedback back into the Intent Lake, and increasingly governs AI-assisted implementation.</p><blockquote><strong>Human responsibility shifts from ownership to stewardship.</strong></blockquote><p>Generative AI initially participates primarily as a downstream consumer of intent. As AI-assisted discovery and analysis mature, it will increasingly contribute to producing and refining intent under human stewardship.</p><h3>A Persistent Context for People and AI</h3><p>The Intent Lake becomes much more than a shared engineering repository. For product engineering it plays the same role that global Markdown files, system prompts, and organizational context play for today’s AI systems - providing persistent, trustworthy context. The difference is that the Intent Lake scales this idea beyond individual AI interactions into an organizational infrastructure with explicit stewardship, governance, and contributions from many disciplines.</p><blockquote><strong><em>Persistent context for people. Augmented context for AI.</em></strong></blockquote><p>For people, the Intent Lake provides a persistent, reliable context for understanding the product domain that survives individual projects, teams, and technology generations.</p><p>For generative AI, the Intent Lake provides augmented context from which specialized agents derive trustworthy, evidence-based solutions rather than reconstructing understanding from isolated prompts and fragmented documents.</p><p>For decades, product organizations have improved how they discover, design, build, and deliver products.</p><p>Generative AI is compressing the traditional engineering flow while shifting the center of gravity from implementation toward explicit intent.</p><p>The next opportunity therefore lies further upstream — not in replacing people with AI, but in creating the infrastructure through which people steward explicit intent and AI transforms it into specialized solutions.</p><p>The Intent Lake provides a common point of reference that guides this transition. It enables product organizations to incrementally establish producer, curator, and consumer roles across disciplines while creating a shared foundation that naturally supports a network of collaborating humans and AI agents.</p><p>Rather than replacing today’s engineering practices, the Intent Lake becomes the organizational north star for evolving toward AI-native product engineering.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be31433d86ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part II - Intent Defines the Solution Space]]></title>
            <link>https://medium.com/@joerg.beringer/part-ii-intent-defines-the-solution-space-b9a6d9a7f635?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/b9a6d9a7f635</guid>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[product-development]]></category>
            <category><![CDATA[product-management]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Wed, 08 Jul 2026 16:06:03 GMT</pubDate>
            <atom:updated>2026-07-08T16:06:03.448Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*72pxhOZd751vouMUuFqkIg.png" /></figure><h4>The second article in a trilogy exploring how AI is shifting engineering from specifying implementations toward operationalizing intent.</h4><p>The previous article introduced three different realms of product development: the existing reality, the documented problem space, and the solution space. For decades, systems and software engineering has connected these realms by progressively translating stakeholder understanding into increasingly implementation-oriented representations. Stakeholder requirements were translated into system requirements. System requirements were translated into architecture, interfaces, user stories, designs, and ultimately systems and software.</p><p>That process reflected the economics of software engineering.</p><p>When implementation was expensive and entirely manual, every downstream discipline required its own view of the future system. Much software engineering therefore evolved around creating increasingly detailed implementation-oriented specifications.</p><p>Generative AI changes that assumption.</p><p>Not because AI understands users or businesses better than humans, but because it increasingly has access to rich representations of the documented problem space and understands them.</p><p>That seemingly small shift changes something much more fundamental.</p><h3>Result-oriented human-AI interaction becomes an engineering principle</h3><p>Anyone who regularly works with generative AI eventually discovers the same lesson.</p><p>The quality of the result rarely depends on instructing on every implementation step. Instead, successful prompts communicate three things remarkably well:</p><ul><li>the intended outcome,</li><li>the relevant context,</li><li>the constraints within which AI is expected to operate.</li></ul><p>Human-Centred AI describes this interaction paradigm as <strong>result-oriented interaction</strong>. Instead of telling software exactly <em>how</em> to solve a problem, we increasingly describe <em>what</em> should be achieved and allow AI to determine an appropriate solution.</p><p>What is happening in software engineering appears to be a logical extension of exactly the same interaction model.</p><p>The interaction pattern does not change.</p><p>It simply moves from individuals to organizations.</p><blockquote><strong>- Individuals communicate intent through prompts.</strong></blockquote><blockquote><strong>- Organizations will increasingly communicate intent through intent-oriented engineering artifacts.</strong></blockquote><p>This is more than a new engineering practice.</p><p>It changes the primary purpose of engineering artifacts.</p><h3>Intent is always paired with constraints</h3><p>One aspect of prompt engineering deserves more attention.</p><p>A useful prompt is rarely just a statement of intent. It combines two complementary elements that together define the space in which AI can operate.</p><ul><li><strong>Intent</strong> describes the destination.</li><li><strong>Guardrails</strong> describe the boundaries of acceptable solutions.</li></ul><p>Those guardrails may include business objectives, regulations, existing platforms, APIs, architectural principles, design systems, performance targets, safety requirements, or organizational policies.</p><p>Intent without constraints produces arbitrary solutions.</p><p>Constraints without intent produce directionless optimization.</p><p>Together they enable purposeful exploration.</p><blockquote><strong>Intent defines the destination. Guardrails define the boundaries of the solution space.</strong></blockquote><p>The same principle naturally scales from prompting an AI assistant to developing an entire product.</p><h3>AI shortens the downstream translation chain</h3><p>Traditionally, stakeholder understanding had to be translated into a sequence of increasingly implementation-oriented representations before software could be built. Each downstream discipline required a representation that matched its own work.</p><p>Traditionally, stakeholder understanding had to be translated into a sequence of increasingly implementation-oriented representations before software could be built. Every downstream discipline translated that understanding into its own conceptual framework:</p><ul><li>stakeholder requirements into system requirements,</li><li>product objectives into user stories and backlog items,</li><li>business rules into interface specifications, APIs, and data models,</li><li>interaction concepts into detailed UI specifications,</li><li>quality objectives into acceptance criteria and test specifications.</li></ul><p>These artifacts were never created for their own sake. They existed because they served as the necessary translation layer between stakeholder understanding and implementation.</p><p>Generative AI begins to shorten that translation chain. Increasingly, downstream engineering artifacts can be derived directly from sufficiently rich intent-oriented representations.</p><blockquote><strong>The bottleneck is no longer implementing solutions. It is making intent explicit.</strong></blockquote><p>That does <strong>not</strong> make system requirements obsolete. Architecture does not disappear, and user stories remain valuable. What changes is their role. They no longer have to serve as the mandatory bridge between stakeholder understanding and implementation. Instead, they become specialized projections of a shared understanding of intent.</p><h3>Intent is more than requirements</h3><p>Intent is often mistaken for another name for requirements. It is much broader than that.</p><p>Traditional engineering artifacts primarily describe the solution itself. System requirements specify capabilities. Interface specifications describe structures. Architectures describe technical organization. Together, they prescribe what the future system shall become.</p><p>Intent operates at a different level of abstraction. Rather than prescribing the solution, it describes the intended outcomes and constraints that should steer it. It captures not only what is necessary, but also what is expected, desired, preferred, constrained, or should be avoided. Stakeholder requirements, business objectives, quality expectations, regulatory constraints, design principles, and many other engineering considerations all become different expressions of intent.</p><p>That makes intent fundamentally different from implementation-oriented specifications. Instead of describing the solution, intent describes the semantic space within which good solutions can emerge.</p><blockquote><strong>Intent does not prescribe solutions. It defines the semantic space that steers them.</strong></blockquote><p>This also explains why intent-oriented engineering artifacts are such a natural input for generative AI. AI is exceptionally good at exploring solution spaces once the steering intent has been made explicit.</p><h3>Interactive systems combine two sources of intent</h3><p>Interactive systems introduce another important characteristic that traditional software engineering has often treated separately.</p><p>Every product organization expresses an organizational intent. It defines why the product exists, which business outcomes it should achieve, which stakeholders it targets, and which constraints must be respected.</p><p>At the same time, every user approaches the system with a human intent. Users have goals they want to accomplish, outcomes they hope to achieve, expectations they bring to the interaction, and mental models that shape how they understand the system.</p><p>Successful interactive systems continuously reconcile these two sources of intent. Organizational intent defines what the product should accomplish. Human intent defines what the product should enable.</p><blockquote><strong>For interactive systems, human intent defines the solution space of the user interface.</strong></blockquote><p>This perspective naturally expands the role of Human-Centred Design. User research is no longer only concerned with understanding people. It becomes the systematic discovery and operationalization of human intent as a first-class engineering input. The engineering challenge is no longer simply to build the system correctly, but to ensure that organizational intent and human intent remain aligned throughout the product lifecycle.</p><h3>From specifying implementations to operationalizing intent</h3><p>The implications reach far beyond prompts.</p><p>If AI increasingly derives specialized solutions from explicit intent, the value of engineering naturally begins to move upstream. For decades, engineering excellence has largely been measured by how precisely organizations specified implementations. Increasingly, it will also be measured by how effectively they operationalize intent at scale.</p><p>This changes the role of engineering artifacts themselves. The primary source of truth gradually shifts from implementation-oriented specifications toward intent-oriented engineering artifacts that describe the objectives, expectations, constraints, and human intentions steering future solutions.</p><p>That shift cannot be achieved simply by adopting another engineering practice. It requires rethinking the engineering process itself. Product organizations must continuously discover, validate, operationalize, curate, govern, and evolve explicit intent throughout the product lifecycle.</p><p>It also creates a fundamentally more collaborative engineering model. Human-Centred Design, Product Management, Engineering, and many other disciplines no longer contribute isolated deliverables. They contribute complementary sources of explicit intent that together become the shared foundation for specialized downstream solutions.</p><blockquote><strong>The value of engineering is moving from specifying implementations toward operationalizing intent at scale.</strong></blockquote><p>That is no longer a prompting problem. It is a fundamental shift in how product engineering transforms upstream understanding into downstream solutions.</p><p>The final article in this trilogy explores the infrastructure that enables this new engineering model.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b9a6d9a7f635" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part I - Requirements Don’t Begin on Paper. They Begin in Reality]]></title>
            <link>https://medium.com/@joerg.beringer/part-i-requirements-dont-begin-on-paper-they-begin-in-reality-58015e02b31e?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/58015e02b31e</guid>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Fri, 03 Jul 2026 06:17:05 GMT</pubDate>
            <atom:updated>2026-07-03T06:17:05.688Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_XL0O_vbQJwP8W0X84dNwQ.png" /></figure><h4>The first article in a trilogy exploring why feature specifications are no longer sufficient to communicate human intent to AI.</h4><p>Over the past year I’ve watched engineering introduce terms such as Spec-Driven Development and Intent-Driven Development. At first I saw them as new software engineering approaches. The more I thought about them, however, the more they seemed like symptoms of a much larger shift.</p><p>As a design leader, I repeatedly found myself contributing to transformational product and platform roadmaps rather than isolated design initiatives. That perspective led me to look beyond the emerging terminology and ask a different question: <em>What larger change are these new approaches actually responding to?</em></p><p>In this trilogy, I’d like to put these symptoms into a broader context that, in my view, reflects a more fundamental shift in how organizations will develop products in the age of AI.</p><p>One of the first visible consequences of that shift is that generative AI has reopened a question many of us thought had already been settled.</p><h3><strong>Do we still need requirements?</strong></h3><p>It is not a new question.</p><p>Over the past decades, requirements have repeatedly been accused of being too expensive, too slow to create, too difficult to maintain, and too quickly outdated. Traditional requirements engineering produced documents that nobody wanted to read. Agile reduced them to lightweight user stories. Lean UX encouraged teams to validate assumptions through prototypes rather than specifying the requirements that lead to them, nor modelling the solution itself. Now generative AI appears to challenge them once again. If AI can generate user interfaces, write code, derive tests, and even propose architectures directly from a well-written prompt, perhaps requirements are becoming obsolete altogether.</p><p>The more I thought about that question, the more I realized it starts with a hidden assumption.</p><p>It assumes we all mean the same thing when we say <strong>requirements</strong>.</p><p>I don’t think we do.</p><p>In fact, I believe the long-running debate about requirements has less to do with documentation practices than with something much more fundamental.</p><p>We have been using the same word to describe different kinds of knowledge.</p><blockquote><strong>The recurring debate has never really been about the value of requirements. It has been about the cost of making understanding explicit.</strong></blockquote><p>That distinction matters.</p><p>Because before we ask whether requirements are becoming obsolete, we first have to understand what they actually represent.</p><h3><strong>Three flavors of requirements</strong></h3><p>When people from UX, Product Management and Engineering discuss requirements, they are often referring to three different things.</p><p>Not different levels of detail.</p><p>Different kinds of reality.</p><p><strong>Need: </strong>Something that already exists in the real world (and is often not met)<br>Related terms: <em>User goals, business goals, operational constraints, regulations, context of use</em></p><p><strong>Stakeholder Requirement: </strong>An explicit deduction from an identified need that the future system is expected to support<br>Related terms: <em>User requirements, desired outcomes, product objectives, context-of-use descriptions</em></p><p><strong>System Requirement: </strong>A description of the characteristics the future system must exhibit to satisfy one or more stakeholder requirements<br>Related terms: <em>Requirements, quality requirements, data models</em></p><p>This distinction is easy to overlook because all three are commonly called <em>requirements </em>and used in an inconsistent way among many other inconsistent synonyms.</p><p>Yet they belong to very different parts of product development.</p><h3><strong>Requirements don’t originate in documents.</strong></h3><p>Human needs do not come into existence because somebody starts a software project. A nurse does not suddenly need a safe medication workflow because a product team creates a backlog. A logistics company does not invent operational constraints because somebody writes a Product Requirements Document.</p><p>The realities that matter already exist. Our first task is not to specify them. It is to understand them.</p><p>That is why Human-Centred Design begins with observing work, interviewing stakeholders, understanding context of use, and identifying user needs that are often implicit rather than explicitly stated.</p><p>Only after these realities have been understood do we begin documenting our understanding of them.</p><blockquote><strong>Stakeholder requirements do not invent needs. They make the implications of existing user needs for the solution explicit.</strong></blockquote><p>System requirements serve a different purpose. They no longer describe reality. They describe the capabilities the future system must provide.</p><p>The difference is not what these artifacts describe, but where they belong. Needs belong to reality. Stakeholder requirements belong to our understanding of reality. System requirements belong to our understanding of the future system.</p><p>Once this distinction becomes visible, many long-standing discussions suddenly make sense. UX has traditionally focused on discovering and understanding user needs. Product Management has translated that understanding into stakeholder requirements. Engineering has transformed those into system requirements that specify to-be- implemented system capabilities and behavior.</p><blockquote><strong>The debate was never about requirements themselves.</strong><br><strong>It was about confusing different realms of requirements.</strong></blockquote><p>Everyone was talking about “requirements,” but they were not talking about the same thing. UX was concerned with discovering existing user needs, Product Management with expressing their implications for the solution, and Engineering with specifying the capabilities of the future system. They were working in different realms while using the same vocabulary.</p><h3><strong>Three realms of product development</strong></h3><p>Looking at product development from this perspective reveals three distinct realms.</p><p><strong>Existing Reality: </strong>The world the product is intended to support<br><em>Human activity, work practices, business strategy, regulations, context of use</em></p><p><strong>Documented Problem Space: </strong>Explicit representations of the realities relevant for the product<br><em>Stakeholder requirements, desired outcomes, product objectives, context descriptions</em></p><p><strong>Solution Space: </strong>Explicit specifications of the intended solution<br><em>System requirements, architecture, designs, APIs, implementation, tests</em></p><p>These realms also explain two terms that are frequently used but rarely defined.</p><p>The activities that help us understand and represent reality before solutions are designed belong <strong>upstream</strong>. Research, product discovery, requirements engineering, and product strategy all operate primarily in this part of the process.</p><p>The activities that transform this understanding into a working product belong <strong>downstream</strong>. Architecture, interaction design, implementation, testing, and deployment all operate here.</p><p>For decades, requirements engineering gradually shifted its center of gravity downstream. The primary engineering artifact became the system requirement, even though the underlying requirements originate upstream in the context of use.</p><p>That made perfect sense while implementation itself was the dominant engineering challenge. Engineering connected the three realms through a sequence of increasingly implementation-oriented representations, allowing solutions to be communicated with sufficient precision for humans to build them.</p><p>Generative AI changes that assumption. It no longer necessarily requires the intermediate translation of human intent into increasingly detailed feature specifications. Instead, it can increasingly reason directly from explicit representations of the problem space.</p><blockquote><strong>Feature specifications were the optimal language for humans to communicate implementations to other humans. They are no longer necessarily the optimal language for communicating human intent to AI.</strong></blockquote><p>This observation raises a more fundamental question. If AI can increasingly reason from explicit representations of the problem space, are many of today’s system requirements actually feature specifications — a translation layer that became necessary because humans had to communicate implementations to other humans?</p><p>This changes the economics of implementation and requires us to rethink how we move from upstream understanding to downstream implementation.</p><p>What happens when AI increasingly understands explicit descriptions of the problem space instead of requiring increasingly detailed descriptions of the solution?</p><p>That question marks the beginning of the next part of this trilogy.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=58015e02b31e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Innovation begins at the user need — not at the feature idea]]></title>
            <link>https://medium.com/@joerg.beringer/innovation-begins-at-the-user-need-not-at-the-feature-idea-ad5374f8e28f?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/ad5374f8e28f</guid>
            <category><![CDATA[user-needs]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[vibe-coding]]></category>
            <category><![CDATA[product-innovation]]></category>
            <category><![CDATA[user-research]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Tue, 10 Mar 2026 11:40:56 GMT</pubDate>
            <atom:updated>2026-03-12T14:12:07.442Z</atom:updated>
            <content:encoded><![CDATA[<p><em>by Thomas Geis &amp; Joerg Beringer, ProContext Inc.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d6N5tcxNNelntFi1XyBt8A.png" /><figcaption>Context Lineage Across Design Stages</figcaption></figure><p><strong>Innovation begins at the user need — not at the feature idea</strong></p><p>Many product discussions start with ideas.</p><p>“What if we added AI to this?”<br> “What if we redesigned the workflow?”<br> “What if we built a smarter dashboard?”</p><p>Ideas are seductive. They feel creative and forward-looking.</p><p>But most breakthrough products do not begin with ideas.</p><p>They begin with something much quieter: a moment where reality reveals that something essential is missing.</p><blockquote>Innovation begins when we observe the gap between what people intend to accomplish and what reality allows them to do.</blockquote><p>Ethnographic observation and contextual interviews makes that gap becomes visible.</p><p><strong>Where innovation actually begins</strong></p><p>Consider a gastroenterologist performing a colorectal cancer screening with a colonoscope.</p><p>At first glance, the procedure appears straightforward: a camera is inserted into the intestine to inspect its inner surface.</p><p>In reality, the task is far more complex.</p><p>The intestine forms loops, curves, and sections with varying diameter. Navigating it is comparable to exploring a cave system. Some areas are easy to inspect. Others are hidden behind folds.</p><p>During observation and contextual interviews of such procedures, a subtle tension becomes visible.</p><p>Even experienced physicians sometimes hesitate:</p><p><em>Have all areas actually been inspected?</em></p><p>Overlooking even a small region can have serious consequences for the patient.</p><p>The physician intends to inspect the entire intestine. But the anatomy of the body makes it difficult to know whether that goal has truly been achieved.</p><p>That moment — where intention meets uncertainty — is where innovation begins.</p><blockquote>The most valuable insights in user research appear <br>where intention and reality diverge.</blockquote><p><strong>From observed context to user need</strong></p><p>Ethnographic observation and contextual interviews reveal real-world behaviors and frictions: actions, statements, workarounds, and moments of hesitation.</p><p>Individually, these are simply data points.</p><p>Their meaning emerges when they are interpreted within the <strong>context in which the task occurs</strong>.</p><p>When multiple observations point to the same difficulty, a pattern appears. That pattern reveals what the user actually needs in order to succeed.</p><p>In the colonoscopy example, one need becomes clear:</p><pre>The gastroenterologist needs to know <br>which areas of the intestine have not yet been inspected <br>in order to ensure that all potential anomalies are detected.</pre><p>Notice what this statement does <strong>not</strong> contain.</p><p>It does not describe a feature.<br>It does not prescribe a technology.</p><p>It simply describes what the user must <strong>know </strong>in that situation in order to achieve an intended outcome.</p><p>That is the moment where innovation potential becomes visible.</p><blockquote>User needs are not feature ideas.<br>They are signals that essential outcomes can be underserved.</blockquote><p>Ethnographic observation and contextual interviews with physicians reveal these needs by <strong>capturing real-world behavior in the context where success or failure unfolds.</strong></p><p><strong>From need to requirement</strong></p><p>User needs are not only functional. They can also be emotional and personal, reflecting ambitions, discomfort, or broader life goals. In both cases, a user need highlights an opportunity.</p><p>But product teams cannot build opportunities. They build systems.</p><p>This is where user requirements enter the picture.</p><p>User requirements translate a need into a statement describing what a system must enable for the user.</p><p>From the identified need above, a user requirement for a future colonoscope system might be formulated as:</p><pre>With the system, the gastroenterologist must be able to recognize <br>which areas of the intestine have not been inspected yet, <br>in order to detect all potential anomalies.</pre><p>This transformation is critical.</p><p>A user need reveals <strong>innovation potential</strong>.<br> A user requirement makes that potential <strong>actionable for design and engineering</strong>.</p><p>Once the requirement exists, many solutions become possible:</p><ul><li>coverage visualization</li><li>spatial inspection maps</li><li>navigation guidance</li><li>AI-assisted anomaly detection</li></ul><p>The solution remains open.</p><p>But every design option can now be evaluated against the same underlying need.</p><blockquote>Needs reveal the opportunity.<br>Requirements make the opportunity designable.</blockquote><p><strong>The hidden lineage behind good products</strong></p><p>Behind every well-designed product lies a traceable chain of reasoning:</p><p>Observed context<br> → implied user need<br> → derived user requirement<br> → implemented system behavior</p><p>When this lineage is preserved, product decisions remain grounded in real situations.</p><p>A user requirement can always be traced back to the exact moment in which the need appeared — the subtask, the difficulty, the observed hesitation.</p><p>Anyone on the team can understand <strong>why the user requirement exists</strong> and which situation it addresses.</p><blockquote>Good product decisions are rarely arbitrary.<br>They are traceable.</blockquote><p>This lineage transforms research from storytelling into <strong>engineering input</strong>.</p><p><strong>Why this chain often breaks</strong></p><p>In many organizations, research results are shared as raw material: interview recordings, transcripts, or large collections of notes.</p><p>While rich in detail, this approach places a heavy burden on product teams. Few engineers or designers have time to watch hours of research footage or read entire transcripts.</p><p>Researchers therefore condense the material into summaries, highlight lists, or presentation slides.</p><p>But each compression step removes part of the original context. The reasoning behind insights becomes harder to reconstruct.</p><p>The lineage between observation, need, and requirement disappears.</p><p>And product teams respond with a familiar question:</p><p><em>“What’s actually new here?”</em></p><blockquote>Observation alone creates knowledge.<br>Structured context creates insight.</blockquote><p><strong>Structuring context with knowledge graphs</strong></p><p>One way to preserve the lineage is to represent the context of use structurally.</p><p>Instead of treating observations as isolated notes, the context can be modeled as a <strong>knowledge graph</strong> connecting:</p><ul><li>stakeholders</li><li>tasks and subtasks</li><li>goals</li><li>pain points</li><li>observed situations</li></ul><p>Within such a structure, relationships between observations become visible. When multiple observations converge on the same difficulty within a task, a user need emerges naturally.</p><p>That need can then be linked directly to the task where it was observed and translated into a user requirement.</p><p>The result is a navigable context with needs and requirements where anyone in the team can move seamlessly from a requirement back to the situation that revealed it.</p><p>Context and derived requirements become <strong>explorable rather than summarized</strong>.</p><p><strong>Scaling the lineage with AI</strong></p><p>Building and maintaining this structure manually requires significant effort.</p><p>AI systems can dramatically accelerate the process.</p><p>Product Context Analyzer (PCA) is an AI-powered system that transforms contextual information into a structured representation of the context of use — including stakeholders, tasks, goals, and observed difficulties.</p><p>From this structure, PCA identifies implied user needs and derives corresponding user requirements.</p><p>The lineage remains intact because each requirement remains linked to the context in which the need appeared.</p><p>The process can begin in two ways.</p><p>Teams can start from a simple description of a situation and let PCA discover the context. Or they can upload transcripts from their own research and allow PCA to transform field observations into the same structured representation.</p><p>Either way, the result is a traceable chain linking:</p><p>observed situations → identified user needs → derived user requirements.</p><p><strong>Toward spec-driven generative design</strong></p><p>Once user requirements are structured and traceable, something powerful becomes possible.</p><p>They become <strong>machine-readable specifications</strong>.</p><p>Instead of treating research insights as informal inspiration, they become explicit constraints that can guide design systems and generative AI.</p><p>In a spec-driven design workflow, structured user requirements can serve as the foundation for:</p><ul><li>automated concept exploration</li><li>generative UI systems</li><li>requirements validation across design alternatives</li><li>traceability between design decisions and real user context</li></ul><blockquote>Generative design does not start with prompts.<br>It starts with <strong>user</strong> <strong>requirements grounded in context</strong>.</blockquote><p><strong>The future of user-centered product development</strong></p><p>User research has never been about collecting stories about users.</p><p>Its real purpose is transforming contextual observations into structured needs and derived user requirements that guide design and engineering.</p><p>Knowledge graphs can represent the context.<br> User needs reveal innovation opportunities.<br> Requirements translate those opportunities into design constraints.</p><p>AI can scale the entire chain.</p><p>In this future, systems like PCA are not simply research tools.</p><p>They become part of the <strong>infrastructure of modern product development</strong> — continuously translating context into structured requirements that feed design, engineering, and generative systems.</p><p>Researchers increasingly become <strong>context analysts</strong>, continuously supplying the product team with structured representations of the context of use and the requirements that emerge from it.</p><blockquote>Ethnographic thinking remains human.<br>Ethnographic observation and context modeling become powered by AI.</blockquote><p>AI can discover context, structure it, and surface emerging needs at a scale no research team could achieve manually.</p><p>Human researchers remain responsible for framing the context, interpreting meaning, and ensuring that innovation remains grounded in real human situations.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ad5374f8e28f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Ethnography Has Left the Field]]></title>
            <link>https://medium.com/@joerg.beringer/ethnography-has-left-the-field-f14de0d6e190?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/f14de0d6e190</guid>
            <category><![CDATA[ai-in-ux-research]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[ux]]></category>
            <category><![CDATA[vibe-design]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Thu, 19 Feb 2026 23:29:55 GMT</pubDate>
            <atom:updated>2026-02-19T23:29:55.974Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EvQwe6gBDulsJM0M9sID9w.png" /></figure><p><strong>And Product Context Analyzer Is the Tool That Makes It Possible</strong></p><p>For decades, ethnographic research was constrained by physics. Researchers had to travel to sites, observe people in their environments, document behavior, and then spend weeks consolidating notes into something coherent. The process was immersive and often revelatory. It surfaced nuance, workarounds, and lived realities that surveys and dashboards could never capture.</p><p>But it was intrinsically limited.</p><p>Most research initiatives begin with strategic concerns: adoption is lower than expected, users do not recognize value, workflows feel fragmented, journeys break down. These questions are legitimate. They frame intent at the business level. Yet none of them are directly observable.</p><blockquote>You cannot observe a strategy.<br>You can only observe a person trying to get something done.</blockquote><p>Usability itself is defined concretely: a specific user achieving a specific goal within a specific context of use. That definition is structural. If research claims to inform usability, it must translate abstract concerns into observable units of behavior:</p><pre>[User] performing [task] in [context]</pre><p><strong>What Field Research Really Struggled With</strong></p><p>Field ethnography remains one of the most powerful research methods available. Observing real work in real contexts reveals tacit knowledge, environmental constraints, and cultural adaptations that rarely surface elsewhere.</p><p>But field research always faced two structural constraints.</p><p>The first was scale. Even the most ambitious programs could observe only a limited number of roles, tasks, and contexts. Entire responsibility structures, edge cases, and contextual variations inevitably remained outside the frame.</p><p>The second constraint was synthesis, and this was often underestimated.</p><blockquote>Collecting observations was difficult.<br>Consolidating them into a coherent representation was harder.</blockquote><p>After the visits ended, the real labor began. Researchers transcribed recordings, clustered findings, reconciled contradictions, mapped journeys, constructed personas, and interpreted patterns. They stitched fragments of lived reality into artifacts that could guide design and strategy.</p><p>This consolidation required enormous cognitive effort. And because it required so much effort, it was episodic. Research became a project — something conducted, delivered, and archived..</p><p>The resulting reports often felt comprehensive. The immersion created confidence. The artifacts appeared holistic. But structurally, they were always partial.</p><blockquote>Depth creates the illusion of completeness.</blockquote><p>Immersion provides intensity, not coverage. Observing selected slices of reality does not produce a continuously explorable model of context. Once the research phase ended, the representation of context rarely evolved at the same pace as the product or the market.</p><ul><li>Strategy changed</li><li>Features evolved</li><li>Contexts of Use shifted</li></ul><p>The constructed context model remained static.</p><p><strong>Ethnographic Logic Still Applies — But the Medium Has Changed</strong></p><p>Ethnography, at its core, is not about physical presence. It is about disciplined decomposition grounded in observable behavior within a relevant context.</p><p>A broad business question becomes researchable only when it is expressed as:</p><p><strong>[Specific user] performing [specific task] in [specific context].</strong></p><pre>[Specific user] performing [specific task] in [specific context]</pre><p>From that anchor, ethnographic reasoning expands outward. What steps are involved? Who is performing these steps? What decisions occur? What constraints shape behavior? What artifacts mediate action? Where do breakdowns emerge?</p><p>This logic does not disappear in the age of AI. It becomes even more essential. The shift lies elsewhere:</p><blockquote>What has changed is not the logic of ethnography.<br>What has changed is where that logic can now be executed.</blockquote><p>Large language models are emerging as a new environment for scalable ethnographic discovery.</p><p>Large language models contain vast amounts of latent knowledge about how tasks are performed across domains, embedded in documentation, manuals, forums, workflows, and support cases. What they do not provide automatically is structure. Without boundaries, they generate narratives. With boundaries, they support decomposition.</p><p>This is where the shift becomes tangible.</p><p><strong>Product Context Analyzer: Ethnography Inside an LLM</strong></p><p>Product Context Analyzer is an AI tool that operationalizes ethnographic decomposition inside large language models.</p><p>It begins with a bounded observable unit of human activity:</p><pre>&gt; A first-time installer setting up the system in a noisy, shared workspace</pre><p>From that starting point, it decomposes the task into subtasks, environmental constraints, dependencies, and role interactions. It surfaces relevant knowledge within that defined context and stitches these elements into a structured knowledge graph — an explorable model of the context of use.</p><p>The result is not a summary — it is a context representation.</p><blockquote>Field research samples fragmented reality.<br>Structured AI decomposition models a comprehensive context of use.</blockquote><p>Because this modeling is virtual and immediate, research is no longer constrained to isolated phases. It can deepen vertically into subtasks and edge cases, or expand laterally across roles, responsibilities, and journeys.</p><p>Engineers can examine micro-actions and task object models. Designers can explore task transitions and breakdown points. Strategists can zoom out to responsibility structures and jobs-to-be-done; all grounded in the same underlying context model.</p><p>The connective tissue between strategy and execution becomes structural rather than rhetorical.</p><p><strong>AI for Acceleration — or as Infrastructure</strong></p><p>A common application of AI in research today focuses on acceleration. Teams use it to summarize interview transcripts, cluster qualitative data, or generate reports more efficiently. These applications are valuable. They reduce manual effort and shorten synthesis cycles.</p><p>But they do not change the structure of research.</p><p>If the underlying approach still relies on sampling limited slices of reality and consolidating them into static artifacts, faster synthesis does not resolve the scalability problem. It simply produces partial models more quickly.</p><p>The limitation remains: context is reconstructed from fragments and then frozen.</p><p>The structural shift occurs when AI is used not as a summarization layer on top of field data, but as an engine for constructing and maintaining the context model itself. When AI participates in decomposition, task expansion, and the stitching of roles and constraints into a unified structure, research moves beyond accelerated reporting toward scalable modeling.</p><p>That difference fundamentally expands what user research can deliver.</p><p><strong>Context First. Field Research Second.</strong></p><p>Instead of beginning with broad exploratory visits and hoping patterns emerge, teams can start using Product Context Analyzer to generate a coherent model of context of use. They can then decompose tasks, map responsibilities, surface constraints, and identify likely breakdown points before entering the field.</p><p>Field research then becomes intentional rather than expansive. It is used to validate critical assumptions, stress-test task structures, and search for anomalies within clearly defined boundaries.</p><blockquote>Modeling context does not eliminate field research. It reorders it.</blockquote><p>When the context of use is already structured, fieldwork becomes sharper. Researchers know which responsibilities matter, which transitions are fragile, and which constraints deserve scrutiny. Instead of mapping reality from scratch, they refine and enrich a living model.</p><p>Ethnography becomes more precise.</p><p><strong>Context as fuel for AI Design</strong></p><p>AI-assisted design and development are accelerating rapidly. Generative systems can produce interfaces, flows, and even applications from high-level prompts. The productivity gains are undeniable.</p><p>The structural risk lies elsewhere.</p><p>When generation is driven primarily by intent or imagination — without explicit requirements grounded in context of use — the output may look coherent while remaining disconnected from real task structures and constraints.</p><blockquote>You get plausible design, <br>but you do not know whether it fulfills real requirements.</blockquote><p>An explicit context model changes that equation.</p><p>When Product Context Analyzer produces a structured knowledge graph of tasks, roles, constraints, and dependencies, it creates more than a research artifact. It creates a context model specification that can be used to inform design decisions and serves as explicit requirements for generative AI design tools.</p><p>Instead of prompting, “Create an admin dashboard,” teams can ground the generation in explicit task structures, user needs, and decisions to be made. The same model can also be used for validation by testing generated design against user stories derived from the same context model.</p><blockquote><strong>As-is</strong>: AI generates plausible UI without context and encourages context-free solutioning. Reality later reveals whether it meets real user needs.</blockquote><blockquote><strong>To-be</strong>: An explicit context model directs AI generation towards purpose-fit solutions and enables immediate validation against inherent requirements.</blockquote><h3>Where Context Works Best</h3><p>Context modeling expands what can be discovered before entering the field. By surfacing tasks, constraints, responsibilities, and dependencies, an explicit context model enables teams to articulate requirements and guide design prior to observational research.</p><blockquote>AI can model context.<br>Only reality validates behavior.</blockquote><p>A more robust approach is to assess AI-generated solutions against explicit requirements derived from the context model — evaluating whether designs support identified tasks, constraints, and user stories.</p><p>Stress-testing in reality should not be confused with replacing users through anthropomorphized AI representations or simulated interviews. Plausible opinions are not behavioral evidence. Real-world validation evaluates solutions against explicit requirements and observed behavior.</p><p>The implication is a clear sequence:</p><ul><li><strong>Model context first through AI-enabled context discovery</strong></li><li><strong>Validate against context requirements second</strong></li><li><strong>Stress-test in reality third</strong></li></ul><p>This sequence preserves the foundational logic of ethnography while radically modernizing how it is practiced.</p><p><strong>Ethnography as Infrastructure</strong></p><p>Ultimately, ethnographic thinking remains the foundation: define observable human behavior, decompose tasks, surface constraints, and separate observational facts from interpretation. What has changed is not the logic of ethnography, but how and where that logic can be executed.</p><p>Where observation once served as the primary gateway to context discovery, context modeling now enables this work to occur continuously and at scale. Field research becomes more focused, design becomes requirement-driven rather than assumption-driven, and strategy becomes anchored in modeled context rather than narrative alignment.</p><blockquote>Strategies cannot be observed directly, <br>but they can be grounded in the context model within which they must operate.</blockquote><p>Ethnography has not disappeared; it has expanded beyond the field as its only home. When context of use becomes explicit, explorable, and continuously expandable, user research ceases to be a phase and instead becomes an enduring capability.</p><p>In this sense, ethnography evolves from episodic investigation into infrastructure — a living context model that informs discovery, guides design, and sharpens validation across the entire product lifecycle.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f14de0d6e190" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Making User Research Methods Align Instead of Compete: A Context-First Perspective]]></title>
            <link>https://medium.com/@joerg.beringer/when-research-methods-compete-instead-of-align-a-context-first-perspective-196425595ecd?source=rss-ba38a7cfefbf------2</link>
            <guid isPermaLink="false">https://medium.com/p/196425595ecd</guid>
            <category><![CDATA[generative-ai-tools]]></category>
            <category><![CDATA[user-research]]></category>
            <category><![CDATA[ux-research]]></category>
            <category><![CDATA[user-experience]]></category>
            <category><![CDATA[product-design]]></category>
            <dc:creator><![CDATA[Joerg Beringer]]></dc:creator>
            <pubDate>Tue, 16 Dec 2025 06:20:32 GMT</pubDate>
            <atom:updated>2025-12-18T16:32:01.623Z</atom:updated>
            <content:encoded><![CDATA[<p>This article reflects on our journey toward a context-first model of qualitative user research, shaped while evolving Product Context Analyzer as a multi-method research companion. It explores why method-driven approaches often result in fragmented outputs and introduces a two-phase research model that separates context discovery from method-specific deliverables.</p><h3><strong>The Challenge of Scaling User Research</strong></h3><p>Qualitative user research has traditionally been organized around <strong>discrete methods</strong>: interviews, task analyses, contextual inquiries, personas, user stories, and scenario-based frameworks. Each approach provides value, but when applied across large organizations, several challenges arise:</p><ul><li><strong>Fragmentation</strong>: Different teams produce isolated artifacts, leading to inconsistent interpretations.</li><li><strong>Duplication</strong>: Similar questions are often explored independently, consuming unnecessary resources.</li><li><strong>Method bias</strong>: Outputs are influenced by preferred methodologies rather than actual user context.</li><li><strong>Limited scalability</strong>: Traditional approaches are labor-intensive and struggle to keep pace with team growth.</li></ul><blockquote>Most research methods are simply different lenses on the same underlying context. Without structuring that context first, outputs often conflict or duplicate effort.</blockquote><p>The key insight is that <strong>context of product use is the true foundation of actionable insight</strong>. Without it, outputs from different methods often conflict or fail to scale effectively.</p><h3><strong>A Two-Phase Research Model</strong></h3><p>To address these challenges, Product Context Analyzer implements a <strong>two-phase approach</strong>:</p><p><strong>Phase 1: Context Discovery</strong></p><p>In this phase, <strong>AI and large language models (LLMs)</strong> automatically identify and structure the factual context in which a product is used. This includes users, tasks and goals, task objects, constraints, pain points, and interdependencies. The output is a <strong>structured, unbiased representation of the product context</strong>, serving as the reference for all subsequent analysis.</p><p><strong>Phase 2: Method-Specific Projection</strong></p><p>Once the context has been discovered and structured, Product Context Analyzer applies established research methods as projections to transform this foundational context into method-specific outputs. This ensures that every established approach — including <strong>Jobs-to-Be-Done (JTBD)</strong>, <strong>Personas</strong>, <strong>Task Analysis</strong>, <strong>User Story Mapping</strong>, <strong>Cognitive Analysis</strong>, and <strong>Scenario-Based Design</strong> — produces deliverables that are <strong>consistent and aligned</strong> with the discovered context.</p><blockquote>Well-structured context is the true foundation of meaningful research; method-specific deliverables are simply perspectives projected on that foundation.</blockquote><p>This relationship between a shared context model and multiple method-specific outputs is illustrated in Figure 1, where context discovery forms the stable center from which different research deliverables are derived.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*wfCEyiLlt-ck4YiH-foWHg.png" /><figcaption><strong>Figure 1:</strong> Context Discovery at the center, surrounded by method-specific outputs. Each output is derived from a method-specific projection of the same foundational context.</figcaption></figure><p><strong>Connected User Research Outputs</strong></p><p>A central insight is that <strong>method-specific outputs are projections of the same factual foundation</strong>. Each method emphasizes particular aspects:</p><ul><li><strong>Jobs-to-Be-Done (JTBD)</strong>: Focuses on user goals and desired outcomes.</li><li><strong>Personas</strong>: Represents user archetypes for design empathy.</li><li><strong>Task Analysis</strong>: Maps procedural steps, dependencies, and critical paths.</li><li><strong>User Story Mapping</strong>: Organizes work from user goals into product features.</li><li><strong>Cognitive Analysis</strong>: Examines decision-making and cognitive load.</li><li><strong>Scenario-Based Design</strong>: Models interaction sequences and contextual constraints.</li></ul><p>By anchoring each method in a shared context, methods stop competing because they no longer project divergent realities. Instead, their outputs remain <strong>complementary, reliable, and reproducible</strong> — avoiding the inconsistencies typical in traditional research workflows.</p><p><strong>Operational Benefits</strong></p><p>Figure 2 illustrates how user input flows through context discovery into method-specific content synthesis, creating an iterative loop where structured context continuously informs research outputs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*D_FXqn1o8hVskn63rNFR4g.png" /><figcaption><strong>Figure 2:</strong> Workflow diagram showing initial user input → context discovery → method-specific deliverables.</figcaption></figure><p>Applying the two-phase model in practice has produced several key benefits:</p><ol><li><strong>Scalability</strong>: Teams maintain a coherent understanding of product context while generating multiple types of deliverables.</li><li><strong>Alignment</strong>: Cross-functional teams reference the same structured context, avoiding conflicting interpretations.</li><li><strong>Efficiency</strong>: Automating context discovery frees researchers to focus on <strong>analysis and strategy</strong>.</li><li><strong>Flexibility</strong>: The same context supports multiple method-specific research deliverables without redundancy.</li><li><strong>Iterative Refinement</strong>: Outputs reveal gaps in context, prompting continuous improvement.</li></ol><blockquote>By separating context from method, user research becomes an operational capability rather than a series of disconnected outputs.</blockquote><p><strong>Observations from Early Adaptors</strong></p><p>Early adopters illustrate how context-first research changes team operations in practice:</p><ul><li><strong>Speed</strong>: Method-specific outputs were produced in <strong>minutes instead of months</strong>, dramatically shortening the decision cycle.</li><li><strong>Cross-team alignment</strong>: All stakeholders operated from a <strong>shared understanding</strong>, improving confidence in insights.</li><li><strong>Knowledge transfer</strong>: New team members could quickly grasp prior research through the structured context.</li><li><strong>Agile experimentation</strong>: Researchers could explore multiple methods without duplicating work, enhancing adaptability.</li></ul><p>The context-first approach acts as a <strong>scaffolding for user research</strong>: it structures, amplifies, and scales human insight while maintaining methodological integrity.</p><p><strong>The Science Behind Context Discovery</strong></p><p>The two-phase model reflects established research principles:</p><ul><li><strong>Contextual Design</strong> emphasizes understanding users, tasks, and workflows prior to design.</li><li><strong>Ontology-based Knowledge Representation</strong> allows explicit modeling of relationships among users, tasks, task objects, and resources.</li><li><strong>Large Language Models (LLMs)</strong> enable scalable secondary research, extracting factual information from large data sources to support context mapping.</li></ul><p>The combination bridges <strong>human-centered methodology and AI-assisted automation</strong>, creating a reproducible, scalable foundation for research operations.</p><p><strong>Implications for Product Teams</strong></p><p>Structuring research context first delivers tangible organizational benefits:</p><ul><li><strong>Consistency Across Methods</strong>: Multiple methods coexist without contradiction.</li><li><strong>Rapid Adaptation</strong>: Updates to product or market context propagate across outputs seamlessly.</li><li><strong>Operational Knowledge</strong>: Context becomes an organizational asset, enabling distributed teams to make consistent, data-informed decisions.</li><li><strong>Reduced Redundancy</strong>: Eliminates siloed research efforts and duplicated analysis.</li></ul><blockquote>Organizations adopting this model reduce redundant user research, maintain alignment, and produce reliable outputs rapidly.</blockquote><p><strong>Our Learnings</strong></p><p>These experiences emphasize that the <strong>value of user research lies in structured understanding, not isolated artifacts</strong>. By capturing context explicitly:</p><ul><li>Teams gain agility to explore multiple methods.</li><li>Outputs are interoperable and reusable across projects.</li><li>Organizational learning accelerates through shared knowledge.</li></ul><p>AI-assisted research, as applied here, <strong>scales human insight while preserving methodological rigor</strong>, transforming research from an isolated activity into an operational capability embedded in product development.</p><p><strong>Implications for Research Operations</strong></p><p>As product organizations grow in complexity and scale, traditional artifact-focused research may struggle to keep pace with evolving markets and cross-functional collaboration.</p><p>The two-phase model offers a pathway forward: <strong>structure knowledge first, generate outputs second, iterate continuously</strong>. By formalizing context, organizations can:</p><ul><li>Seamlessly integrate multiple research methods</li><li>Maintain speed and consistency</li><li>Focus human effort on interpretation and high-value analysis</li></ul><p>This evolution demonstrates how user research can move from isolated projects into a <strong>scalable, repeatable operational capability</strong>, enabling teams to make informed design decisions efficiently and confidently.</p><blockquote>While the tools and technologies may evolve, the underlying principle remains constant: meaningful user research starts with a shared, structured understanding of context.</blockquote><p><em>Written by Joerg &amp; Thomas, creators of Product Context Analyzer.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=196425595ecd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>