Engineering has plenty of AI — what it needs is coordination
The next step for AI in engineering is to connect specialized agents throughout the product-development lifecycle
- By:
- Jens Krueger
Speaking with automotive and manufacturing engineering leaders, we often hear two things. First: “We already have several AI pilots running.” And shortly after, almost always: “But somehow, they’re not scaling.”
That’s not a coincidence, and it’s not because of the wrong tools or a lack of ambition.
It’s a structural problem: As products get more complex, the way engineering teams coordinate work across requirements, architecture, software, testing and compliance isn’t keeping pace.
Products are becoming more software-defined and connected, while regulatory demands are growing. Architectures and requirements span mechanics, electronics and electrical systems, software and services. At the same time, verification, validation and compliance are becoming more complex and dynamic.
The processes, methods and tools behind engineering have evolved over decades to protect quality and maintain control. But they’re fragmented, and few were designed with AI in mind.
So, how do we build a new coordination layer for complex product development?
Why isolated AI use cases only get you so far
AI is already helping engineering teams analyze requirements, generate test cases, search technical information and support software development. It’s improving consistency and giving engineers more time to spend on key activities.
In automotive and manufacturing, we’ve implemented these solutions for requirements engineering, engineer-to-order automation, intelligent document processing and technical self-service, among others. And yet, local efficiency doesn’t translate into end-to-end performance.
When capabilities remain isolated — one tool extracting requirements, another supporting software teams and another generating test cases — they create exactly the problems we’re trying to solve: manual handovers, inconsistent data, missing traceability and recurring coordination loops.
Solve enough individual tasks and the next bottleneck appears: coordinating them throughout the product-development process.
Agentic product development: Turning assistance into orchestration
In agentic product development, instead of one generic assistant, multiple specialized agents work together. One supports requirements analysis while another interprets architectural dependencies, and others help with software development, verification, validation, compliance, change management and project management.
The goal is not to replace engineering judgment. Your teams benefit when AI prepares work, connects information, checks consistency and speeds up handovers. People remain responsible for review, approval and decisions with major technical or regulatory impact.
Making it work requires three things:
- Orchestration controls which agent becomes active and when, which data it needs, which tool it uses and when a person must intervene.
- Agent skills package engineering knowledge into reusable, validated instructions: processes, methods, templates, examples, tool usage and quality criteria.
- The Model Context Protocol (MCP) allows agents to interact with external systems in a controlled way. For engineering teams, MCP helps move AI beyond the chat interface and into integrated engineering workflows.
3 workflows where agentic AI makes a difference
The biggest gains come when agents connect work throughout the development lifecycle. Three workflows illustrate the opportunity.
1. From requirements and architecture to verification and validation
Requirements and architecture define the quality of many downstream steps. When they are incomplete or interpreted too late, they create risks that affect development, testing and compliance.
In an agentic workflow, a new specification or request for quotation can trigger a coordinated chain:
- A requirements agent extracts, classifies and evaluates requirements.
- An architecture agent interprets dependencies and interfaces in the system model.
- A verification and validation agent derives test methods, test cases and coverage information from this context.
SysML v2 can play an important role here. Beyond modeling the system, it can give agents a machine-readable representation of that system — including its requirements, relationships and interfaces. That gives different agents a shared context for working across engineering tools and disciplines.
As a result, the systems engineering and development model becomes less dependent on manual document handovers. Instead, teams can connect requirements, architecture, testing and validation through traceable, model-based workflows.
2. Specification-driven software development
Software-defined products need closer connections between specification, implementation, testing and documentation. While that may sound obvious, in practice, it’s often not.
Specification-driven software development uses requirements and design intent as executable context for software teams. Agents can translate specifications into tasks, generate code, prepare tests, update documentation and trigger validation steps in continuous integration and continuous delivery (CI/CD) pipelines.
Connecting specifications, code, tests and documentation in one workflow exposes inconsistencies earlier, gives software teams better context and reduces information lost in handovers.
In safety- and quality-critical environments, human responsibility remains essential. AI speeds up preparation, analysis, generation and validation but doesn’t replace technical approval.
3. Continuous compliance
Engineering organizations often treat compliance as a downstream documentation task. This model is becoming less effective.
Regulations, industry standards, cybersecurity requirements and market-specific rules need to be considered early in requirements, designs, tests and release decisions. When this happens too late, it creates rework, delays and risk.
Continuous compliance uses AI to identify regulatory relevance earlier, connect requirements with engineering artifacts, maintain traceability and prepare evidence more efficiently. This is not automated certification, and it shouldn’t be. The value lies in earlier impact detection, better evidence management and greater transparency.
NTT DATA offers tools such as Syntphony Compliance Management for homologation, conformity, registration and certification-related processes.
The data foundation: Agents need shared meaning
Multiagent systems are only as good as the data foundation they rely on. When requirements, architectural elements, test cases, code, regulations, project plans and change requests are locked in isolated tools, reliable context is missing. Agents may find data but don’t necessarily understand its meaning.
A semantic layer gives systems and agents a shared definition of concepts such as products, requirements, functions, interfaces, components, tests and defects. Only when everyone means the same thing can agents coordinate reliably.
Microsoft Fabric and Fabric IQ serve as platforms for knowledge graphs, ontologies and semantic enterprise models. More important than the specific technology, however, is the insight that scalable agentic AI is not purely a model question. Scaling it also depends on data, semantics, integration, and the transformation of processes, methods and tools.
Without shared meaning, AI remains confined to isolated tasks. With it, agents can connect workflows across tools and disciplines.
Using Systems Engineering Copilot as an accelerator
We’re positioning our Systems Engineering Copilot — based on Microsoft Azure and designed for productive use in enterprise environments — as a reusable accelerator for complex product development.
It’s not a tool for a single systems-engineering method. Our approach targets agentic workflows in existing engineering landscapes — for example, for requirements management, document analysis, test-case generation and tool integration.
The advantage is that teams don’t need to start from scratch for every use case. They can build on reference architectures, integration patterns, CI/CD, infrastructure as code, repositories, connectors and agent skills.
Systems Engineering Copilot is located within NTT DATA’s broader engineering capability: We combine processes, methods and tools consulting with data and AI expertise, reusable engineering assets and support for taking solutions into production.
How to get started
Target key workflows that can realistically be improved and which have the data and ownership you need to proceed:
- Discover and assess: Bring engineering, IT, data and business teams together to identify the three engineering workflows with the highest coordination effort and the greatest potential. Agree on the use cases, key performance indicators (KPIs) and priorities. Determine where requirements, architecture, software, compliance, verification and validation are sepatrated.
- Define, build and evaluate: Define a proof of concept with real data, clear KPIs and human-in-the-loop oversight. Use it to validate selected use cases with real data and engineering context. Based on Systems Engineering Copilot, we follow an efficient approach covering infrastructure setup, implementation, KPI evaluation and roadmap planning.
- Scale: Move successful proofs of concept into an industrialized delivery model with DevOps, data engineering, enterprise architecture, reusable components, change management and governance. Scale successful use cases using our AI Factory model.
Governance is a prerequisite for scaling. When agents gain access to data, tools and workflows, you need clear rules: human approval, security, observability, cost control and lifecycle management.
Benefit from scalable, value-focused agentic workflows
Agentic product development is the creation of a new coordination layer — one that will help you turn AI capabilities into scalable, controlled and value-focused engineering workflows.
That means looking beyond the next AI pilot and asking: Where does coordination slow down engineering today?
Our AI portfolio for engineering helps you connect the people, processes, data and tools behind those workflows and prove what works in practice.
What to do next
Explore how NTT DATA helps manufacturers and product-driven organizations modernize their core operations and engineering environments to speed up AI adoption.
Author
Jens Krueger
Jens Krueger is Executive Consultant and Engineering Competency & Offering Owner for AI in Engineering at NTT DATA, where he shapes the global strategy and portfolio for AI-enabled engineering transformation in the automotive and manufacturing industries. With more than 30 years of experience in product lifecycle management, systems engineering, enterprise architecture and technology consulting, Jens advises clients on how to modernize R&D and product-development processes, navigate increasing software and systems complexity and unlock value from AI throughout the product lifecycle. Combining deep industry expertise with a strong background in strategy and IT, he works with organizations worldwide to translate emerging technologies into scalable transformation initiatives. His current focus is helping engineering leaders build the capabilities, operating models and technology foundations required for the next generation of AI-driven product development, improving engineering efficiency and operational excellence.