Bringing Generative AI to the Shop Floor: End-to-End UX Design for Siemens Industrial Copilot
4 min read · 774 words
Designed end-to-end UX/UI for the Siemens Industrial Copilot plugins and agents interface, from concept through launch-ready screens and data-layer specification.
The Challenge
Factory engineers operate in high-stakes environments where interrupted workflows carry real costs. Their primary tool, TIA Portal, is a deeply embedded, complex system built around decades of industrial automation logic. Asking engineers to adopt a generative AI layer on top of that — one capable of orchestrating multiple AI agents and plugins simultaneously — meant the design had to meet them where they already worked, not redirect them to somewhere new.
Siemens needed a copilot experience that felt native to the shop floor context: precise, trustworthy, and fast to navigate under pressure. The challenge wasn't simply adding a chat interface to an enterprise product. It was designing an interaction model that made AI agent orchestration legible and controllable for engineers who have zero tolerance for ambiguity or unexpected system behaviour.
The absence of established design patterns for industrial AI copilots made this harder. There were no direct precedents to borrow from. Every structural decision — how agents surface, how plugins are invoked, how TIA Portal project context flows into the conversation — had to be reasoned from first principles against the realities of factory floor use.
Key Decisions
- 01
Designing for Existing Workflow, Not Workflow Replacement
The copilot was positioned as an augmentation of TIA Portal, not a standalone product. This meant anchoring the interaction model to the engineer's current mental model — project context, device hierarchies, existing task states — rather than asking them to context-switch into a separate AI environment. The trade-off was constraining the design space: some generative AI affordances that work well in consumer or knowledge-work tools had to be deprioritised because they would have introduced unfamiliar patterns into an already demanding environment.
- 02
Making Agent Orchestration Visible and Controllable
Generative AI systems that operate through multiple agents create a specific UX risk: the user loses track of what is acting, why, and on what data. The decision was made to design explicit visibility into agent states and plugin invocations — surfacing which agent was active, what context it was drawing on, and where in a task sequence the system currently sat. This added interface complexity but was treated as non-negotiable given the industrial stakes of miscommunication or unintended automation.
- 03
Specifying the Data Layer Alongside the UI
Rather than treating the data-layer specification as a handoff concern for engineers, it was incorporated into the design process directly. This decision ensured that the interaction model was grounded in what the system could actually surface — preventing UI promises that the backend couldn't fulfil and reducing the risk of late-stage redesigns. The trade-off was a broader scope for a single designer, but the result was a tighter, more coherent specification that development teams could act on without significant interpretation.
- 04
Prototyping Across Fidelity Levels Before Committing to High-Fidelity
Low-fidelity prototypes were used to stress-test the structural logic of the copilot experience before any visual execution began. This decision was made because the interaction model — agent orchestration, plugin management, TIA Portal context integration — was genuinely novel and needed validation at a conceptual level before committing design effort to pixels. The risk of skipping this step would have been discovering structural problems deep into high-fidelity work, at far greater cost.
The Solution
The end-to-end UX/UI design covered the full product journey from initial concept through to launch-ready screens. This included concept documentation that defined the interaction model and design rationale, low-fidelity prototypes used to validate structural decisions, and high-fidelity UI that met Siemens' production standards for the Industrial Copilot plugins and agents interface.
A data-layer specification was produced alongside the UI, giving development teams a clear, designer-authored account of how information needed to move through the system to support the designed experience. The deliverables were scoped for handoff without requiring significant interpretation — screens, documentation, and specification were designed to function as a coherent package.
The Results
Siemens received a complete, launch-ready design system for the Industrial Copilot plugins and agents interface — one of the first UX implementations of generative AI agent orchestration in an industrial automation context.
The concept documentation and data-layer specification gave cross-functional teams a shared reference that reduced ambiguity at the design-to-development boundary. The interaction model established a clear, controllable pattern for how factory engineers engage with AI agents and plugins within their existing TIA Portal workflow — without disrupting the operational logic they depend on.
The work delivered a credible, production-ready foundation for Siemens to bring generative AI capability to the shop floor in a form engineers could trust and act on.