# The Age of Intent: From MCP Prototype to Organizational Transformation > Designing for intent rather than for interfaces: what a conversational flight-booking prototype built on MCP reveals about the road to transformation in 2026. - Date : 2025-10-24 - Lecture : 8 min - Catégorie : genai - Tags : Generative AI, Age of Intent, MCP, Conversational Design, Agentic AI - URL : https://www.adservio.fr/en/insights/articles/l-ere-de-l-intention-du-prototype-a-la-transformation ## TL;DR - The announced "death of the interface" is not an extinction event but an evolution: intent is becoming the new interface between users and systems. - A conversational flight-booking prototype built on an MCP server showed that building a demo has become trivial, transforming your systems to act on intent has not. - Designing for intent does not remove the UI: it becomes generative, assembled on the fly to show only what serves the user's goal. - Four lessons for organizations: map intents, design APIs able to handle ambiguity, prototype on a real use case, and participate in agent ecosystems rather than trying to control them. - Succeeding at this shift is above all an organizational and cultural challenge, not just a technological one. ## Intent Is Becoming the New Interface This article is the second in our Age of Intent series, exploring what it takes to move from building prototypes to transforming organizations, in an intent-driven future. What happens when we stop designing interfaces and start designing for human intent? When I declared that "the interface is dead," it was not a prediction of extinction, it was a call to evolve. In 2026, as generative AI, multimodal agents and adaptive interfaces transform how people interact with technology, intent has become the new interface. Users no longer navigate screens: they express goals. And systems must understand those goals, reason about them and act in real time, embracing the ambiguity inherent in human language. For enterprises, the stakes go far beyond user experience. When a customer asks their assistant to "find the best flight to Melbourne on Tuesday," it is the assistant that decides which airlines to query, which offers to present and in what order to rank them. Being legible, queryable and reliable for these new intermediaries is becoming a competitive advantage every bit as decisive as ranking on search engines once was. This moment matters because the shift is already operational. AI copilots interpret our requests inside productivity tools. Voice assistants and agentic systems book appointments, draft proposals and execute tasks across ecosystems, often without a screen ever being displayed. Over the past few weeks, we explored what it actually takes to build for this new paradigm. The verdict is unambiguous: building a prototype has become trivial. Transforming your systems and your organization to act on intent absolutely has not. ## A Flight-Booking Prototype Built on an MCP Server To explore what an intent-first world might look like, we built a simple flight-booking prototype. Using fictional data and a lightweight Python Model Context Protocol (MCP) server, the open standard that lets a language model connect to real data sources and functions, now adopted across all major AI platforms, we wired it to a React interface. The prototype exposed three tools to the model: flight search, fare lookup by date and availability checks. Each tool was described in the MCP server with its input and output schema, and that alone was enough for the model to decide when to call it, with which parameters, and how to combine the results into a genuinely useful answer for the traveler. ### A Few Hours to a Working Conversational Experience Within a few hours, we had a working conversational experience. It was exhilarating: a few lines of code, a quick API integration, and users could type natural-language requests like "show me flights from Sydney to Melbourne." This is exactly what the MCP ecosystem has made routine: where custom integration once took months, an MCP server exposes tools that any frontier model can discover and invoke, with native handling of permissions and context. The lesson of this first stage is twofold. The technological barrier to entry has collapsed: any team can now assemble a credible conversational experience in a week. But that ease is deceptive, it shifts the difficulty toward everything a demo never has to face: real data, edge cases, security, scale, and accountability when the agent gets it wrong. > Related read: [The Model Context Protocol: Beyond the Hype, the Standard for AI Agents](https://www.adservio.fr/en/insights/articles/le-protocole-model-context-au-dela-de-la-tendance): Architecture, 2026 specification, official registry, security: how the Model Context Protocol went from hype to the production standard for AI agents. ## When Complexity Kicks In: Ambiguity, Context and Memory And then the complexity began. In a traditional web form, users fill in every field up front, origin, destination, dates, passengers. A conversational interface does not work that way. Users start with a goal ("I need to get to Melbourne") and clarify details later ("Tuesday," "with two kids"). The system has to handle ambiguity, remember context and update its understanding dynamically, turn after turn. ### Handling Ambiguity Without Frustrating the User In our case, the language model handled contextual memory, recalled earlier messages and asked for clarification like "did you mean next Tuesday or this Tuesday?". The frontier models of 2026 excel at this disambiguation exercise, and context engineering, deciding which information to keep, summarize or discard at each turn, has become a discipline in its own right. But what surprised us most was not how the model handled the conversation. It was how much it changed the interface itself. This management of state is the real engineering challenge. Which elements should persist across turns, the destination, the budget, the household composition? Which ones expire, a date that was considered and then discarded? A system that forgets frustrates the user; a system that remembers everything ends up reasoning over noise. The prototype taught us that the boundary between the two gets settled use case by use case, not in the abstract. ## The Interface Is Not Dead: It Is Becoming Generative We found that users still needed to see the trip holistically, both the outbound and return legs at once, and to adjust details visually, much as they would in a traditional booking engine. The difference: they did it with less data displayed and fewer steps. Instead of listing dozens of prices, the interface could show a daily calendar with an average or minimum fare, surfacing only what was relevant to the expressed intent. ### Show Less, but Show What Matters In other words, designing for intent did not eliminate the UI, it reshaped it. The intelligence was not only in the backend; it lived in the decision of what to reveal, when, and how to best support the user's goal. This is the whole logic of the generative interfaces going mainstream in 2026: UI components assembled on the fly by the system based on the detected intent, rather than fixed screens defined in advance. Conversational experiences thus challenge decades of design assumptions: users express goals, not inputs, and the system must infer, reason and orchestrate what comes next. This generative interface also imposes a trust requirement: the user must understand why the system is showing these particular options and must be able to take back control at any moment. The best designs of 2026 make the inference visible, "I am showing you morning flights because you mentioned a 2pm meeting",instead of hiding it behind opaque magic. > Related read: [The interface is dead. Welcome to the era of intent](https://www.adservio.fr/en/insights/articles/l-interface-est-morte-place-a-l-ere-de-l-intention): OpenAI Dev Day reveals the next phase of generative AI: a unified ecosystem where conversation, commerce, and creativity coexist. The dawn of intentional interfaces. ## Four Lessons from the Field for Organizations Based on our work, four lessons stand out for organizations preparing for this shift, lessons that apply well beyond the flight-booking case. These lessons are not theoretical: they directly condition an organization's ability to exist in tomorrow's channels. A product catalog an agent cannot read, an API that demands twelve mandatory fields before answering, a booking system with no conversational state, each is an invisible wall between your offer and a customer who has nevertheless expressed the intent to buy it. ### Map Intents and Design APIs That Think First lesson: map intents, not just journeys. Identify what your customers are trying to accomplish, not just where they click, journeys describe yesterday's how, intents describe tomorrow's why. Second lesson: design APIs that think. Audit your APIs and data products for conversational readiness. Can they handle partial inputs? Provide contextual responses? Expose their capabilities through a properly governed MCP server, with production-grade permissions and traceability? ### Prototype with Purpose and Participate in Ecosystems Third lesson: prototype with purpose. Start small, but make it real, pick a high-impact use case, connect it to live data and observe how users behave once freed from the constraints of a rigid UI. Fourth lesson: prepare to participate, not control. In an intent-first world, your brand will live across ecosystems, conversational assistants, agent networks, voice interfaces, that you will not own. The organizations that expose their capabilities cleanly are the ones agents will choose to call. > Related read: [Agentic AI at Work: How Autonomous Agents Are Transforming Organizations](https://www.adservio.fr/en/insights/articles/l-ia-agentique-au-travail-comment-les-agents-autonomes): Explore how agentic AI systems are revolutionizing the world of work by automating complex tasks and augmenting human capabilities. ## The Road Ahead: An Organizational Revolution The age of intent is not about eliminating interfaces, but about making them invisible when they add nothing, and illuminating when they matter. We are moving from designing paths to designing purpose, from clicking buttons to expressing goals. This is not a UI revolution; it is an organizational one. To succeed, businesses need not just better technology but a new mindset, one that treats customer intent as the starting point for every interaction, and context as the connective tissue that ties everything together, from data to design to decision-making. In concrete terms, this means getting teams that often ignore each other to work together: API owners must think about semantics and discoverability, designers must learn to design systems that decide what to show, and governance leaders must build the tooling for continuous evaluation of agent behavior. It is this collective effort, far more than the choice of any particular model, that will determine who succeeds in the age of intent. Generative AI has made understanding language trivial. Acting on intent, however, demands consistency across data, design and governance, which is precisely the subject of the third installment in this series, dedicated to organizational readiness. The tools are here. The possibilities are real. The question is: are we ready to build for intent, and not just for interaction? ## FAQ ### What does it mean that "intent is becoming the new interface"? Users no longer interact with screens and forms, they express goals directly ("I need to get to Melbourne on Tuesday"). It is up to the system to understand that intent, handle ambiguity and orchestrate the steps needed to fulfill it. ### Did the flight-booking prototype eliminate the user interface? No. The conversational experience simplified input, but users still needed to see their trip holistically (outbound and return, a fare calendar): the intelligence shifted toward deciding what to reveal and when, which is the principle behind generative interfaces. ### What role does the MCP protocol play in this kind of prototype? The Model Context Protocol is the open standard that lets a language model discover and invoke real data sources and functions. It has made building a conversational prototype trivial: a lightweight MCP server is enough to expose tools that any frontier model can call, with permissions and traceability. ### What are the main lessons for an organization preparing for the age of intent? Four lessons stand out: map users' intents rather than just their journeys, design APIs able to handle partial inputs and respond with context, prototype quickly on a real, high-impact use case, and accept that the brand will need to exist within agent and assistant ecosystems it does not fully control.