Beyond the Prompt: Why AI-Assisted Engineering Requires a New Foundation of Trust

From AI Capabilities to Engineering Confidence

Artificial intelligence is rapidly moving from experimentation into mainstream engineering practice. Requirements can now be generated, test cases can be proposed automatically, and engineering artefacts can be analysed at a scale that was previously impossible. Across industries such as defence, aerospace, automotive, and medical technology, software-intensive systems are becoming increasingly dependent on AI-assisted workflows.

The challenge is no longer whether AI can contribute to engineering activities.

The challenge is whether we can trust the outcomes.

For organisations developing mission-critical and safety-related systems, trust has never been derived solely from the competence of an individual engineer, a software tool, or increasingly, an AI model. Trust is established through evidence. It comes from demonstrating how a requirement evolved into an implementation, how a decision was made, how a defect was detected, and how compliance obligations were verified.

As AI becomes part of the engineering process, the industry must rethink how that evidence is generated, maintained, and assured.

The Limits of Traditional Review

Historically, software assurance has relied heavily on review processes. Requirements reviews, architecture reviews, code reviews, and testing activities provide confidence that a system behaves as intended.

AI introduces a new challenge

Modern engineering teams are increasingly using AI to draft requirements, propose test cases, identify duplicate specifications, analyse change impacts, and assist engineers throughout the lifecycle. PTC's roadmap for AI-enabled engineering includes requirement quality analysis, semantic search, test generation, traceability assistance, change impact analysis, risk mitigation and engineering assistants embedded across the lifecycle:

• While these capabilities can improve productivity, they also create new assurance questions:
• What evidence exists to support an AI-generated recommendation?
• How can engineers validate AI-assisted decisions?
• What traceability exists between AI outputs and engineering intent?
• How can organisations demonstrate compliance when AI is involved in safety-critical decisions?

Traditional review techniques alone cannot answer these questions.

Trust Comes From Traceability

In high-integrity environments, confidence is established by maintaining a continuous chain of evidence.

This is where Application Lifecycle Management becomes critical.

ALM systems have traditionally provided the backbone for requirements management, verification, validation, risk management, change control and traceability. In software-defined products, they enable organisations to connect requirements, architecture, implementation, testing, risk assessments, and compliance evidence within a single managed environment.

The emergence of AI does not reduce the importance of traceability.

It increases it.

If AI assists with requirement authoring, then those requirements must remain connected to downstream verification artefacts. If AI recommends a test case or identifies a risk, organisations must be able to understand the context from which that recommendation was generated and how it influenced subsequent engineering decisions.

Traceability becomes the mechanism that transforms AI outputs into auditable engineering evidence.

Software-Defined Defence and Mission-Critical Systems

These challenges are particularly significant within defence and other regulated sectors.

Software-defined systems continue to grow in complexity as products combine software, hardware, systems engineering, cybersecurity controls, and increasingly autonomous capabilities. As complexity grows, disconnected engineering disciplines create risk.

PTC's integrated engineering approach recognizes that software, systems, and product engineering can no longer operate in isolation. Digital Product Traceability establishes connections between requirements, tests, designs, changes, and product configurations across the engineering lifecycle. This enables organisations to maintain visibility and governance while managing increasingly complex software-defined systems.

For mission-critical programmes, assurance depends on maintaining continuity across engineering domains. Requirements cannot live independently from designs. Designs cannot be separated from verification activities. AI recommendations cannot exist outside established governance frameworks.

The engineering lifecycle itself becomes the assurance mechanism.

AI Must Operate Within Governance Boundaries

Trustworthy AI requires more than technical accuracy.

It requires governance.

PTC's AI strategy emphasizes governed lifecycle data, centralized intelligence services, observability, identity controls, and controlled execution as foundational elements of AI adoption. AI maturity is described as progressing from "Advise" to "Assist" and ultimately "Automate", with increasing levels of governance and oversight applied as responsibility increases.

Similarly, PTC's cybersecurity framework identifies responsible AI governance, AI steering committees, secure development practices, and controls against emerging threats such as prompt injection, adversarial manipulation, data leakage, and model abuse. AI-generated code is subjected to the same Secure Software Development Lifecycle controls as traditionally developed software.

This approach highlights an important principle:

AI should not become an exception to engineering governance.

It should become part of it.

The Next Stage of Software Assurance

The future of software assurance will not be built around proving that AI is always correct.

That goal is neither realistic nor achievable.

Instead, organisations must focus on demonstrating that:

• engineering decisions remain traceable;
• risks remain visible;
• compliance obligations remain verifiable;
• AI contributions remain explainable;
• governance controls remain enforceable.

In other words, assurance shifts from validating isolated outputs to validating the engineering system that produces them.

This is where ALM becomes more strategic than ever.

Rather than serving solely as a requirements repository, modern ALM provides the framework through which organisations establish confidence, manage evidence, maintain traceability, and demonstrate trustworthiness across increasingly AI-assisted engineering environments.

Conclusion

As AI becomes embedded across the engineering lifecycle, organisations developing high-integrity systems face a fundamental challenge: how to maintain confidence in increasingly automated processes.

The answer is unlikely to come from AI models alone.

Confidence will be derived from governed engineering processes, robust lifecycle management, secure-by-design development practices, and comprehensive traceability linking requirements, decisions, verification activities, and outcomes. The future of trustworthy AI in engineering will therefore be built not only on smarter algorithms, but on stronger evidence. And that evidence will be created, connected, and governed through the digital thread enabled by modern ALM.