AI Wiring Harness Engineering in the Software-Defined Vehicle

Introduction and industry context

Competitive pressure, geopolitical shifts and the move to climate neutrality and e-mobility are shifting value creation from classical mechanics into electronics and software architectures. The wiring harness (vehicle electrical system) comprises all electrical conductors in main and secondary runs. In premium vehicles it has become one of the heaviest, most expensive and error-prone assemblies, with over 2,500 individual wires in the main run alone in some cases.

Dependence on manual production was exposed during the Ukraine crisis: failures of highly specialised harness plants (including Kromberg & Schubert in Zhytomyr) led to Europe-wide production stops. The industry is therefore pushing automation and reshoring. In parallel, the term harness engineering has become established in software development: the systematic orchestration of AI, context engineering, architectural constraints and DevOps pipelines for autonomous AI agents. The dual meaning (physical wiring harness and digital software harness) reflects the convergence of both worlds.

Evolution of vehicle electrical system architecture

12V vs 48V and zonal architecture

12V has been the standard since the 1960s. Today, high power demand from sensors, actuators and infotainment on 12V leads to heavily oversized copper cables. The answer is the 48V zonal architecture: for the same power, current drops to a quarter, and cross-sections and weight fall sharply (e.g. from about 273 g/m to about 27 g/m for comparable function). That improves range and product carbon footprint.

Topology is shifting from function-oriented (domain-based) to geographic (zonal). In zonal architecture, one zonal controller per vehicle sector drives all sensors and actuators; they communicate over a central high-speed backbone (e.g. Automotive Ethernet) with a central compute unit. This cuts physical complexity and enables AI-supported topological optimisation and algorithmic routing.

AI in ECAD platforms and generative design

Developing a modern vehicle electrical system requires simultaneous consideration of electrical parameters (voltage drop, signal integrity), mechanical constraints (bend radii, packaging), thermal hotspots and manufacturing aspects. AI supports through generative design, algorithmic routing and semantic analysis.

Siemens Capital and EDA AI

The Capital suite (Capital Essentials, Capital Logic Designer, Capital Systems Integrator) is a central MBSE platform for automotive and aerospace. The new EDA AI system integrates generative and agent-based AI, including with NVIDIA NIM and Llama Nemotron. Agents handle layout optimisation, thermal simulation and verification. In harness engineering, Capital automatically generates production-ready drawings, BOMs, NC files and manufacturing reports and computes voltage drops and sneak paths in real time. The Siemens Industrial Copilot (Hannover Messe 2024) provides engineering support in natural language.

Zuken E3.series and Harness Builder

Zuken E3.series provides object-oriented data management with bidirectional sync of 3D geometry and 2D schematics. The Harness Builder automatically converts the 3D-routed harness into a scale 2D formboard drawing (flattening), calculates bundle diameters and dimensions protection sleeves, grommets and labels. With Harness Builder 2025 and E3.series 2026 come ERP and cloud MCAD integrations and exports for Telsonic ultrasonic and OMA braiding machines. The AI Copilot in E3.series automates formboard design and answers textual prompts (e.g. fault finding in the project) in seconds.

Graph-based platforms: SPREAD.ai and Manex AI

SPREAD.ai (Berlin) converts mechanical, electrical and software information into a graph-based functional model. The Product Explorer allows navigation of complex architectures; the Error Inspector matches fault patterns to dependencies in the graph. One European OEM was able to fix faults 60% faster with SPREAD and save over 20 million euros in costs. Manex AI (Munich, 2023, 8M euro seed) focuses on the engineering/quality interface: manufacturing optimisation agents for root-cause analysis and dynamic rerouting in global production networks.

Requirements engineering and AI

The start of every vehicle electrical system project is requirements engineering (RE), traditionally document-heavy. AI and large language models (LLMs) analyse unstructured requirements, uncover contradictions and translate specifications into structured form. SPREAD.ai offers a Requirements Manager that links incoming requirements (e.g. from RFQs) with product architectures, variants and cost data and enables carry-over parts and cost estimates in minutes. Tools such as Aqua AI, Innoslate (GPT-based RE assistants) or Copilot4DevOps turn requirements into BDD scenarios (Gherkin). Whiteboarding tools like Miro or Mural also use AI to derive flowcharts and architecture sketches.

Data standards and the digital twin

Vehicle Electric Container (VEC)

Via VDA 4968 / PSI 21, the Vehicle Electric Container (VEC, version 1.2) was established as a tool-independent digital product model. It strictly separates document version and part version and covers topology, connectivity, part master data, variant control and mechanical metadata. Only this container allows AI systems and cloud services (e.g. Catena-X) to interpret the design, perform manufacturability checks and generate machine code.

Digital twin over the lifecycle

Design twin (as-designed): the AI-supported MBSE model in Capital or Zuken E3. Production twin (as-built): VEC data merges with MES into a bidirectional digital thread. Operation twin (as-maintained): predictive maintenance and anomaly detection by matching sensor data to the structure model; frameworks such as ROS with AWS or Gazebo enable programmatic simulation.

Harness engineering in software development

In the Software-Defined Vehicle, the physical wiring harness and the digital software harness converge. Martin Fowler coined "harness engineering" in the context of AI-supported software development: the human designs environments, specifies intent and sets architectural guardrails for autonomously acting AI agents (e.g. OpenAI Codex). This includes context engineering, architectural constraints and controlled dependencies.

OpenAI demonstrated in an internal experiment that a small team shipped a beta product with around one million lines of code without writing code manually. Codex agents worked autonomously on pull requests, bug reproduction and validation. Critical was the "harness": machine-readable artefacts, structure tests and enforced layering (types, config, repo, service, runtime, UI). Linters and CI enforced compliance. That is the digital equivalent of topological rule checking in ECAD.

Platforms such as Harness.io and OutSystems provide infrastructure for AI DevOps, CI/CD and low-code. Studies also show risks: many organisations report faster deployments with AI but also concerns about software vulnerabilities and cloud cost from inefficiently generated code. Robust guardrails and security testing remain necessary.

Production: automation and reshoring

The goal is full traceability into physical production. Cables and sleeves are flexible and hard for conventional industrial robots to handle. The Next2OEM consortium (Audi, Kromberg & Schubert, Komax, TE Connectivity, KOSTAL, semantic PDM, ArtiMinds, Bär Automation) demonstrated the first fully automated manufacture and assembly of a wiring harness in Ingolstadt in 2026. ArtiMinds addresses handling with force-torque sensors and reinforcement learning: robots learn from plugging operations and adjust trajectories dynamically.

The process chain is digital: the ECAD model is passed to machines via VEC. Komax/Schleuniger handle cut, strip and crimp; Telsonic ultrasonic and OMA braiding machines receive parameters from Zuken Harness Builder. AGVs (Bär Automation) move parts; route planning is optimised by AI (including Fraunhofer ITWM, IPA). Where full automation is not possible, worker assistance systems are used, e.g. the assembly glove "ClickID" from Voss Automotive with sensors and AI for real-time feedback when connectors click home (zero-defect manufacturing).

Research: KI4BoardNet and Fraunhofer

KI4BoardNet (German BMBF, 37.3M euros, Dec 2022 to Nov 2025), coordinated by CARIAD, brings together 22 partners (including Infineon, NXP, Bosch, COSEDA, TE Connectivity, FZI, Fraunhofer IDMT, TUM, TU Dortmund, RWTH Aachen). The aim is to turn the vehicle electrical system from a passive transport medium into a self-regulating component. The FZI is developing SoCs with RISC-V and hardware acceleration for ML in zonal controllers (edge AI). Concepts such as auction theory for resource management are intended to predict energy demand and control allocation dynamically. Results are to be validated in a vehicle demonstrator by end 2025.

Fraunhofer ITWM offers BordNetzSim3D for real-time simulation of the bending mechanics of cables and sleeves and ML-based prediction of fatigue under thermal and mechanical load. Fraunhofer IDMT (Oldenburg) works on "The Hearing Car" project on AI-based acoustic sensor platforms in the vehicle electrical system: detection of emergency sirens and predictive maintenance by classifying environmental and wear-related sounds, with edge AI and data-sparse transmission (privacy by design).

Regulation and cybersecurity

The European AI Act requires transparency and documentation for AI in safety-critical systems. For vehicle electrical system design, companies must document in a traceable way (explainable AI) what data generative algorithms were trained on and how decisions are made. The Cologne Institute for Economic Research stresses the need for the regulation to remain manageable for mid-sized companies. Cybersecurity is central: AI can introduce vulnerabilities and faulty configurations; closed ecosystems and architectural guardrails reduce the risk. Catena-X and data-trust models developed by Fraunhofer ISST enable secure, decentralised data exchange along the value chain without compromising IP.

Conclusion and strategic implications

The move to zonal 48V high-speed architectures is more than efficiency: it is a prerequisite for automation. Only the algorithmic simplification of the wiring harness in systems such as Siemens Capital and Zuken E3 creates the geometric conditions under which AI-driven robots (Next2OEM) can assemble flexible materials precisely. Full digitisation of the chain, from the AI copilot in ECAD via VEC to intelligent production, removes the primary rationale for offshore manufacturing; reshoring becomes a lever with AI. The semantic bridge between the physical vehicle electrical system and the digital software harness is real: both domains combat uncontrolled complexity through strict architecture, generative agents, continuous checks and an end-to-end digital thread. Graph-based platforms and startups show that data silos are breaking down; linking fault patterns, tolerances and dependencies in the digital twin shortens time to market and reduces SOP risks. With KI4BoardNet, VEC standardisation, LLM integration in RE and AI robotics, Germany holds a strong position in the global technology race.