Our Manifesto of Agency Design

For decades we have built digital products based on a simple and reassuring principle: if the user clicks A, then B. Deterministic flows, mapped paths, user journeys designed to guide, predict and have control over every variable.

Then came AI. And with it, silently, that way of imagining the user experience began to be questioned and new scenarios opened up.

We're not talking about AI as a feature (the automatic suggestion, the intelligent filter, the support chatbot). We are referring to something more radical: systems capable of acting autonomously, interpret the context, make decisions, complete complex objectives without anyone having defined every single step. In short, agentic systems.

Our Manifesto of Agency Design

The expected change in the way we think about UX Design it's almost a small culture shock. During the last AI Festival, we presented the Manifesto of Agency Design than Conflux, which sets out the terms of this passage and lays the groundwork for a new User Experience design.

La agentic design it's not the evolution of interfaces. It's a paradigm shift in how experience takes shape. It's not a matter of new tools or new component libraries. It's not 'adding AI' to an existing product. It is rethinking from scratch what it means to design an experience when the experience itself ceases to be a fixed path and becomes something that emerges as it happens.

From experience as a path to experience as an emerging process

The customer journey has always been the designer's conceptual map. States are designed, transitions are defined, touchpoints are optimized. The work is essentially cartographic: representing the territory of interaction and making it as fluid as possible.

In the agentic world, this map doesn't disappear. But it ceases to be a definite line. The experience it is no longer a fixed path to follow, rather it is a process that emerges, guided by intent, context and designed capabilities.

In concrete terms, this means that two users with the same objective (booking a trip, managing a case, solving a problem) can cross radically different paths, because the agent interprets the specific situation and acts accordingly. The system does not follow a flow: it thinks about a space of possibilities and chooses an action.

A new design responsibility

Letting experience emerge in this way does not mean letting chaos ensue. It's not even leaving everything to the machine, as a more superficial narrative about AI continues to insist. IT IS a new type of project responsibility, more sophisticated and in some ways more demanding than the previous one. Before, we planned every step. Ora Let's design the conditions so that the right steps can emerge. The difference is substantial.

The customer journey becomes probabilistic and adaptive; it's no longer deterministic, but it's not random either. We design the structure that operates at a deeper level, that of intentions, constraints and available capacities.

What changes in the role of design

If experience is no longer a path to be drawn but a process to enable, the role of the designer changes profoundly. The starting point is a distinction that seems subtle, but that's all:

We don't design screens. We don't design closed flows. Let's design conditions.

Conditions are the set of elements that make it possible for an agent system to operate reliably and usefully. Contextualized and quality data, on which the agent can reason without introducing noise or distortion. Clear and well-defined action skills, that the agent can select and combine to achieve a goal.

Borders and guardrails are the limits within which the system can move and that protect the user, the service and the company from unwanted drifts. Stop and escalation criteria, or the rules that define when the agent must stop, ask for confirmation or pass the ball to a human being.

Designing all of this requires skills that go far beyond user interface design or information architecture. It requires systemic thinking, knowledge of AI models and their limitations, the ability to anticipate the side effects of poorly calibrated autonomy.

The design no longer controls every step, but it makes what can emerge reliable. It changes the object of control, not the responsibility. On the contrary, responsibility grows, because design errors in an agent system do not appear as a button in the wrong place. They manifest as unexpected behaviors that are difficult to correct.

Models, Agents, and Error Amplification

In product teams and company boards, there is often a belief that a more powerful AI model will automatically solve problems. All you have to do is choose the right one, integrate it and expect better results. This belief does not correspond to reality. A more powerful LLM doesn't fix a wrong design, it amplifies what it finds: without context, amplifies the noise; without guardrail, amplifies the error.

A model that operates on poor, ambiguous, or uncontextualized data does not simply produce mediocre outputs. The problem is that these outputs seem true even though they are wrong. This is the main criticism of the 'tool-first' approach that still dominates in too many discourses about AI in the company.

The models are not neutral, they amplify design choices. In an Agent AI, this means that a mediocre design doesn't just produce a mediocre experience, it unfortunately produces damage.

Agentic design exists precisely for this reason, and it assumes the responsibility of establishing the conditions on which the action of the AI will be based.

The new coordinates of agentic design

Redefining the role of design also means redefining the fundamental concepts with which we work.

The agent as a new form of interface. For decades, the interface was the screen, the form, the menu. Today, the interface can be an agent: a system that interprets intentions, acts on behalf of the user, returns results. It's not a metaphor. It is a real change in the point of contact between person and service, with enormous implications for trust, perception and control.

The service as a real product. When an agent completes a task (book, send, update, solve) what the user experiences is not an interface. It's a service. The product ceases to be the app and becomes the system's ability to produce value autonomously and reliably. This is a turning point for those who design and for those who measure success.

The context as a new code. In traditional interfaces, system behavior is defined by code. In agentic systems, behavior depends to a large extent on the context: the information available, the instructions received, the reference data. Designing the context (what the agent knows, how he knows it, in what form) is the designer's new act of design.

Orchestration as a new flow. User flow has always been the narrative structure of the product. In multi-step or multi-agent agentic systems, flow becomes orchestration: how the different components coordinate, in what order they act, how they manage conflicts and dependencies. Designing the orchestration is designing the experience.

Not chatbots. Not automation. Something new.

It is necessary to be explicit about what agentic design is not, because confusion on this point is still widespread and leads to poor decisions.

We are not designing chatbots. Chatbots are responding. The agents are acting. The difference is essential and radically affects the way in which the system is designed, tested and evaluated.

We are not automating existing flows. Automation takes a human process and executes it faster. Agentic design does not automate a process, rather it creates the ability to achieve a goal through paths that were not planned in advance.

What we are doing is design systems to which we can delegate. It's something much more complex than simply transferring a task. It means building trust and setting expectations, so we can accept that the system will make decisions in our absence.

For those decisions to be the right ones, an AI agent must know when to act, how to act, and when to stop. Precisely defining and designing these conditions is the designer's job.

A reliable collaboration between people and intelligent systems

The future of the experience won't be drawn in pixels or wireframes. It will be designed as a reliable collaboration between people and intelligent systems. Agentic design requires abandoning established certainties and developing new skills, new mental models, new evaluation criteria.

We don't want to make prophecies. AI agents are already a reality in the most advanced products on the market, and will soon become a new and natural way for people to get what they need and to interact with companies and services.

‍

Conceived and written by Luigi Greco - CEO & Founder Conflux

‍