The era of the interface as a "cover-up" for organizational complexity is ending. With the rise of the agentic, design does not disappear: it shifts from the surface of the screen to the structure of the contract. The click no longer carries authority — semantic clarity does. A case for moving from designing journeys to architecting robust business models.
The graphical interface was a non-aggression pact. A comfortable compromise between the chaotic complexity of our processes and the need to make them vaguely usable. That era is over. The arrival of agentic interfaces is not a mere ergonomic evolution; it is the dissolution of that façade — a truth test that exposes three decades of organizational debt hidden behind buttons and dropdown menus. The question is not how we will speak to the machine, but whether we still have anything coherent to say to it.
The graphical interface (GUI) was the greatest lie of enterprise computing. Not a malicious lie, but a lie of omission, a tacit arrangement. Faced with management systems (ERP, CRM) whose internal complexity reflected decades of mergers, acquisitions and political compromise, the GUI offered a truce. It promised to make the Liferay monster or the SAP labyrinth navigable, on condition that the user gave up trying to understand the underlying logic.
The deal was simple:
"Don't ask why. Click here, then there."
This mediation transformed entire swathes of the white-collar workforce into terminal operators. The skill was no longer the mastery of a craft, but the almost esoteric knowledge of a click sequence.
This dynamic created the "super-users" — those ambivalent figures, both bottlenecks and indispensable pillars, whose power rested not on transferable business expertise but on mastery of a software opacity. The interface was not a tool of emancipation; it was a control mechanism that ensured compliance by anesthetizing critical thinking.
This pact sedimented organizational debt. Rather than simplifying a convoluted billing process, we added a screen, a field, a validation step. Each organizational patch translated into a software wart, cementing complexity a little further.
The Inverted Conway's Law was in full effect: it was no longer communication structures that dictated system architecture, but the rigid architecture of systems that imposed organizational silos and absurd workflows. The GUI was the map of that prison.
The arrival of conversational agents and LLM-driven low-code platforms ends this truce. Salesforce's recent announcement, making all of its objects and processes accessible through APIs and natural-language commands, is not a routine update. It is the admission that the graphical interface has become the main limiting factor. The productivity gain is no longer in click optimization, but in the outright dissolution of the graphical mediation step.
To dialogue with an agent, the unit of exchange is no longer the click but the intent. Trying to phrase a request like "give me the list of at-risk customers in the French SMB segment who have not been contacted by sales for 90 days" immediately exposes the fragility of our definitions.
What is an "at-risk customer"? Is the calculation standardized, or does it depend on an analyst's interpretation of an Excel file? Is the "SMB segment" a stable property or a volatile label? Is a "sales contact" a call logged in the CRM, an email, or a calendar entry?
The agent cannot guess. It executes or it fails. And its failure is a diagnosis. It does not say "the AI doesn't work" — it says:
Your definition of "customer risk" is unusable.
It reveals that what we thought was a process was in fact a ritual — a series of habits held together by the elbow grease of a few individuals. The individual acceleration that AI promises crashes head-on against the viscosity of a system whose grammar nobody truly masters anymore.
The direct consequence is the need for polymorphic interfaces, capable of adapting to each person's cognitive preferences. A manager will prefer dictating a voice request between two meetings; an analyst will build a complex script to automate a report; an operational user will interact via a chatbot embedded in Teams. The idea of a single, standardized interface — a relic of industrial Taylorism — becomes nonsense.
The immediate fear, especially from CIOs and security officers, is anarchy. If everyone can interact with the central system in their own way, how do you guarantee traceability, compliance and maintainability?
How do you debug a problem when the "transaction" is a natural-language conversation? The objection is legitimate, but it rests on a flawed premise: that freedom of interaction implies chaos in execution.
The solution does not lie in restricting modes of interaction, but in the robustness of the manipulated object. This is where concepts from software engineering, like Eric Evans' Domain-Driven Design (DDD), become central to management. The founding idea of DDD is to build a "ubiquitous language" shared by business experts and developers, and to materialize it in clear software objects (the Domain Models). The challenge is no longer to agree on an interface, but to agree on the definition and rules governing a "Customer," a "Contract" or an "Invoice." Once that logical contract is established, the means of accessing it can multiply without risk, because they all hit the same auditable, secured semantic core.
A new risk emerges. The risk of replacing the bureaucracy of the click with the tyranny of the "perfect prompt." We can already see "prompt engineering experts" appearing, along with managers who — to keep control — will want to dictate the exact way to formulate requests, recreating forms and scripts disguised as conversation. That would be repeating the same error, swapping one yoke for another. Power would no longer belong to whoever masters the interface, but to whoever writes the agent's instruction manual.
This temptation is a political reflex, not a technical necessity. It is the final convulsion of the culture of a priori control. Complexity, once justified by software opacity, would be reincarnated in the supposed art of formulation. Yet the goal of agentics is not to find the one right way to ask a question, but to let the system understand an infinity of variations around the same intent.
The counter to this tyranny is not prompt training, but investment in systemic auditability. Control should no longer apply to the upstream action (the click, the prompt) but to the downstream outcome (the transaction, the data change). In a world where interactions are fluid and personalized, the only guarantee is the system's ability to record each operation in an immutable log, to explain why a decision was taken, and to trace each instruction back to its source — whether it came from a human or another agent. We move from the psychological reassurance of the form to the mathematical assurance of the log.
This shift requires a radical overhaul of skills and roles. The UX/UI designer, focused on the ergonomics of graphical journeys, must mutate or become a niche. The critical need is no longer to ease an imposed journey, but to define the semantics of the business domain. The new strategic role is that of "contract architect" — a hybrid profile at the intersection of business, legal and technology.
Their mission: not to draw screens, but to lead the work of conceptual clarification. With teams, they must answer fundamental questions: What are our 50 critical business entities? What invariant rules govern them? What states can they take? Who is allowed to initiate which state transition? This work, which may seem tediously philosophical, is in fact the most concrete and value-creating act in an agentic economy. It is the construction of the bedrock of truth on which all future interactions will rest.
For technical teams, this means a refocus on engineering fundamentals: clean APIs, rigorous domain modeling, and an obsession with observability. For managers, it means moving from a role of task supervisor to that of meaning curator. Their responsibility is no longer to check that people click on the right buttons, but to ensure everyone shares an unambiguous understanding of the concepts they manipulate. It is infinitely more demanding work — and eminently human.
The interface asked us where to click. The agent asks us what we really want. Few organizations are ready to answer honestly.
Analysis
Organizations invested millions to deploy AI. Their teams refuse to use it. This isn't a training problem, or a change management problem. It's a coherence problem: incentives, processes and metrics still reward the very behaviours AI was meant to replace.
Analysis
L'avènement de l'IA agentique rend obsolète l'immense infrastructure méthodologique bâtie par l'industrie Tech depuis vingt ans. Alors que la production de code devient une commodité, le véritable défi des dirigeants n'est plus technologique, mais structurel : il s'agit de démanteler l'encombrant "exosquelette Agile" devenu un frein, pour réallouer massivement le capital financier et humain de l'usine à code vers la stratégie et la découverte de valeur.
Analysis
Integrating Generative AI is more than a tech upgrade; it’s a brutal reveal of organizational entropy. While digital natives operate at the "speed of intent," traditional firms are suffocating under a "viscosity tax", a silent hemorrhage of cognitive bandwidth that threatens their very survival.