State of the Market
The pace of change in engineering has never been faster. Products are becoming more complex, more connected, and under more regulatory pressure than ever before. At the same time, AI use in lifecycle management has expanded from chatbots and copilots into a large set of assisted work capabilities. These developments have made engineering timelines shorter and have raised the expectations for what Application Lifecycle Management (ALM) and Product Lifecycle Management (PLM) tools should be capable of. The main question for engineering firms is how to keep up with the pace of adoption without losing control over governance, compliance, and security.
Problem/Engineering Need
AI has moved beyond isolated tasks, and engineering teams must explain what shaped each output and show how it was checked before it was accepted after using it. The same model should not both produce the work and determine whether it is correct. Engineering firms will need an evidence chain showing what inputs shaped an output, which tools and models were used, how the result was tested, who reviewed it, and why it was accepted.
Agents are an example of how quickly that burden can grow since they can carry out several connected steps before a person reviews the results. This output creates a larger trail of decisions and change history that must be tracked. The hardest errors to catch are often not dramatic failures, but small changes that appear reasonable and produce the wrong result. Human review is still required, but it can benefit from support such as static analysis, regression testing, coverage checks, and other methods to return a repeatable result.
Many organizations still lack the lifecycle coverage needed to preserve this evidence across tools. Teams may be able to see the final artifact without being able to reconstruct the prompts, source data, tool calls, test results, changes, and approvals that produced it. This is especially important in highly regulated industries, where a technically correct output may still be unusable if it was created outside an approved process. Regenerating a test, for example, is not the same as preserving a regression record that shows what changed and whether product behavior was affected.
Vendor Response
Major vendors are continuing to embed AI more deeply into their platforms. IBM made Engineering AI Hub 1.2 available in March. PTC released Codebeamer AI 1.0, with capabilities focused on requirements and test-case assistance. Siemens introduced nine industrial AI copilots across platforms including Teamcenter, Polarion, and Opcenter.
These releases show that the market remains centered primarily on embedded AI assistance rather than broad autonomous engineering agents. Vendors are using AI to make established engineering tools easier to operate, accelerate artifact creation, and support specific lifecycle tasks within defined controls. Broader agent workflows remain largely limited to pilots and narrowly scoped use cases. The longer-term advantage may therefore come less from introducing another standalone chatbot and more from making existing engineering tools accessible through the models and interfaces customers already use. Vendors will also need to support multiple models and deployment environments, since many customers will not permit sensitive engineering data to be processed through public AI services.
Market Gaps
AI is not free to scale. Every prompt, code generation, and validation cycle consumes tokens. A use case that works for one enterprise may not be practical across millions of lines of code, thousands of users, and repeated test cycles that a different enterprise may require. Token use is only part of the whole cost.
The models enterprises build workflows around could also be updated, changed, or discontinued with little warning. In June 2026, the U.S. government ordered Anthropic to suspend access to Claude Fable 5 and Mythos 5 over national security concerns. Senator Mark Warner relayed what the head of the NSA told him: “This tool broke into almost all of our classified systems, not in weeks, but in hours.” That was a controlled test, but the point still stands. AI is advancing fast enough that governments are beginning to intervene directly in what models can be publicly accessed. These models available to the general public are becoming powerful enough to raise serious security questions not just for the government, but for enterprises as well.
The real challenge in AI use in lifecycle management is not adoption alone. Enterprises must balance the speed of adoption with the governance, compliance, and security demands that come with it, while at the same time remaining ready for the technological shifts. Faster output could lead to more defects, especially when the amount of generated work grows faster than the organization’s ability to review it. AI can make an experienced engineer much faster, but it does not give an inexperienced engineer the judgment needed to recognize a bad answer.
The gap is no longer between what AI can generate and what engineers can imagine, but between what it can generate and what an organization can prove.
Key Questions
- How can engineering organizations establish the requirements traceability needed across ALM and PLM systems as AI tools continue to be updated?
- Is a human properly reviewing and approving the work these systems produce?
- How can an organization ensure an agent performs its assigned task accurately while operating under strict regulations and changing technologies?
- Can an organization securely build its workflow around a large language model, or would a smaller in-house model be better suited to the task?
- What are the best practices for preparing current design processes for future changes, and which vendors are best positioned to help engineering organizations maintain compliance?
Stay tuned for more insights and best practices stemming from VDC Strategy’s AI-driven lifecycle management research.
