Generative UI:
for humans and machines
Complexity changes shape
Anyone with a few miles on their tech clock remembers the evolution of interface design: ultra-basic, stark screens, with a few focusing entirely on efficiency. Who still remembers the IBM AS/400?
For decades, enterprise dropdown menus have been treated worse than cluttered utility closets, not to mention forty-column data tables and deeply nested filters in ERP systems. The result? Most users stick to the exact same screens or rely on the same "mnemonic" shortcuts to retrieve information.
I still remember a past study I conducted—replete with metrics—on a legacy software system: it revealed that several features considered "vital" to maintain and develop weren't even opened by users.
Today, the rules have changed. Enterprise applications are no longer consumed solely by people clicking static forms. Working alongside us are autonomous AI agents—synthetic co-workers compiling data, web browsing agents gathering external intelligence, and background process orchestrators—as well as external members of the organization (e.g., consultants, freelancers, clients, etc.).
Expecting a single, rigid interface to serve both a human operator and an intelligent digital agent is like building a dirt road for horse carriages and expecting thousands of high-speed autonomous vehicles to navigate it alongside cars driven at normal speeds by real people.
The chatbot sidebar trap
When companies realized AI integration was non-negotiable, the initial response was a shortcut: stick a chat widget on the edge of the screen.
Ah, the holy "Chatbot".
This Frankenstein approach failed to solve complexity: it simply added noise. Turning your enterprise software into a receptacle where AI is pasted onto the side does not streamline workflows. The user ends up with the same cluttered interface in the center and a side window that has zero awareness of what is actually happening on the page.
The real bottleneck isn't forcing the user to chat with the software. It's radically simplifying the experience by integrating a dual perspective and providing alternative access channels.
Designing for both worlds
A modern Enterprise application must serve several distinct user profiles consuming the exact same interface:
1. The Human Operator: Does not want to wade through fifty submenus. Demands immediate clarity, zero cognitive overload, contextualized information, and the ability to supervise high-risk actions with a single tap (Human-in-the-Loop).
2. The AI Agent: Whether a web browsing agent or an internal digital "co-worker," the agent must "read" and comprehend the interface without misunderstandings, while also benefiting from dedicated access layers (e.g., via MCP, REST APIs, etc.).
3. The Fallback: Transitioning to new systems isn't simple. As I've been emphasizing for a long time, it's essential to complement these new paradigms with fallback screens (essential yet practical) that still ensure full navigability—for instance, through Master-Detail patterns or generalized layouts.
Generative UI: that compose themselves
What is Generative UI? It doesn't mean letting an LLM randomly paint buttons on the screen. It means the frontend framework no longer loads rigid layouts written three years ago. Instead, it interprets user intent (and operational context/state) to compose the exact visual elements required on the fly.
Consider a financial workflow:
- Instead of presenting a dense ERP screen with full balances, vendor histories, and edit forms, the agent analyzes the request and presents a concise card with key information, core metrics, and actionable steps.
- The interface adapts to user preferences and operational context: if the user prefers a summary chart or a compact tabular view, the UI self-composes on the spot.
- The agent presents its underlying reasoning visually, allowing the user to validate or correct with a single click.
- Naturally, a "manual drive" mode must always exist, allowing the system to seamlessly scale down to the fallback layout as mentioned.
Under the hood, the user interface becomes the output of a dynamic schema resolution:
// Conceptual contract for dynamic UI composition
interface AutoComposedCanvas {
userIntent: "REVIEW_VENDOR_INVOICE";
agentContext: {
confidence: 0.94;
flaggedRisk: "DUPLICATE_IBAN_DETECTED";
};
// Context-aware visual components generated on demand
layoutSchema: {
primaryWidget: "RiskAlertBanner";
comparisonView: "SideBySideInvoiceDiff";
allowedActions: ["APPROVE_PAYMENT", "REQUEST_HUMAN_AUDIT"];
};
// Props, comps, data, ..
}In this model, the interface stops being a static wall of buttons and evolves into a living canvas. Stripping away visual noise is the most advanced and effective act of design one can perform in Enterprise software.
Bespoke: where UX meets AI
Applying Generative UI correctly is not a simple graphic exercise or a superficial UI trend. It demands a deep understanding of how data flows from the backend to agents, and from agents to the eyes of those who need it.
This creates a crystal-clear imperative: truly understanding the underlying information. We are no longer designing rigid "flows," but their most fluid, optimal shape. It's about determining which components and data blocks belong together (relationships and affinities), while factoring in accessibility, device context, and the capability of agents to co-work alongside people (agentic workflows & digital co-workers).
Interfaces that also serve agents
Do agents need a separate API, or is the interface enough?
How do you design for humans and machines without building twice?
What does this have to do with accessibility?
Transform your Enterprise user experience
You don't necessarily need to rebuild your entire SaaS product from scratch: perhaps it's simply time to redesign the CX/UX layer rather than just polishing the UI.
Tag: AI, Development
Publication date: August 3, 2026
Last revision: August 3, 2026
