AI as an end user: every screen instantly accessible to agents

In the Wercstat framework, AI acts as a digital assistant: it supports an employee live in the screen, or handles tasks fully independently. Every custom and ERP screen is open to it directly, without separate integrations or APIs having to be built, documented and maintained.

A look under the hood: why this can be so simple.

The Wercstat framework assumes that the UI technology (Angular, React, Vaadin, Vue, Web, Mobile, etc.) changes structurally far more often than the server-side technology (Java, Spring). To make Wercstat future-proof, the UI is fully decoupled from all business logic. This lets the framework move easily with new UI technologies without the underlying business logic having to change.

The behaviour of the UI is handled entirely server-side in Java: which fields are required or editable, which actions are active at which moment, how lists and forms are linked, as well as screen defaults and notifications.

That makes it possible to now add AI to every screen across the whole application with ease. At its core, it is no different from a textual UI.

The framework automatically splits every screen into two interfaces: a visual UI for the human employee and an invisible, digital interface for the AI via MCP (an open standard with which AI agents communicate with software). This allows AI to be deployed within applications in two ways.

Handling tasks independently (headless mode)

The AI logs in in the background and functions as an independent user. It opens screens, reads the data and autonomously processes complex information flows, such as entering purchase orders or updating master data. Because the AI communicates directly with the logic behind the screen, this is fast and stable.

Assisting the user (co-pilot mode)

The AI can also follow along live in the screen an employee is working in at that moment. While the employee enters data, the AI makes suggestions in the background, flags errors or fills in missing data on the spot. The visual UI moves live with the AI’s actions, so human and machine collaborate on the same screen.

Context-driven assistance: domain knowledge

The AI can be equipped per screen with specific knowledge of the field and the sector in which the application runs.

In co-pilot mode, each screen gets its own AI skill, which the customer can then extend further themselves with specific sector guidelines or technical terminology. This way the AI delivers well-founded advice directly in the active screen.
In headless mode, this architecture offers full flexibility for external parties. They can hook into the headless interface of the Wercstat screen with their own, sector-specific skills and knowledge models. This lets the AI carry out complex, sector-specific tasks autonomously within the boundaries of the system.

Security & governance: AI with its own passport

Within Wercstat the AI gets its own account and goes through the same access control as an employee. As a result, this functionality rides entirely on the framework’s built-in security structures.

Roles and rights: the AI gets its own username and account. Through Role-Based Access Control (RBAC), exactly which roles and authorisations the AI has is determined in advance. An AI with limited rights can, for example, enter data but never approve payments.
Strict boundaries: the AI is bound by exactly the same business rules and validations as a human user. The system rejects incorrect or unauthorised input from the AI directly at screen level.
Full audit trail: in headless mode, every action the AI carries out independently is logged under its own username. In co-pilot mode the employee performs the action, so they remain responsible for the change in the audit trail. In both cases every change remains fully traceable for review afterwards.

Available right away

Every screen designed within the Wercstat framework is directly accessible to AI agents. Manually building and documenting separate APIs for AI applications is thereby unnecessary. When a screen is functionally adjusted or renewed, the AI interaction moves along automatically.

Curious what AI as an end-user or assistent means for your business software? Feel free to get in touch.