tion workforce, but it is an effective tool that will facilitate their access to system data. The potentially critical conse - quences of AI hallucinations also necessitate developing and vetting approved DoW governance and AI agents to interface with the model. These agents must be prompted to quality-check all responses against authoritative data and, if the data don’t exist, relay the hard truth in - stead of generating it. In addition to vetted agents, there is potential to connect multiple Large Language Models (LLMs): one to generate the answer and then additional LLM(s) to vet the data and perform quality checks. This again doesn’t eliminate the need for a human reviewer but will increase the quality of AI agent responses. Diagram generation. One power - ful feature of SysML is its ability to communicate complex information using structured, unambiguous dia - grams. AI can reliably parse and edit text files. However, diagrams are built for specific stakeholders and require nuance and judgment on how best to display the information. AI tools can generate images, but it remains to be seen how effectively they can produce useful, accurate, human- readable diagrams. Tool availability. Commercial tools and workflows are not yet fully available for us in DoW acquisition. Multiple tool vendors are moving forward in developing these capa - bilities. Tool vendors are moving forward in developing these capa - bilities, with promising tools already coming to market. Sensmetry offers Syside, which embeds an AI assistant directly into a SysML v2 modeling en - vironment. SysModeler.ai generates model structures from uploaded doc - uments. Siemens has deployed Team - center Copilot, which enables con - versational requirements analysis, and IBM introduced an AI-powered agent for its Rhapsody tool. AI-native platforms, including Celedon Davinci and Dalus.io, have also built prompt-
driven environments on SysML v2’s textual syntax. However, the DoW does not have any AI + SysML v2 tools vetted and available for acquisition programs at this time. Security verification and vetting of powerful tools with embed - ded AI likewise poses a major chal - lenge.
ATS is pushing the workforce to integrate new technology and work - flows. Our current environment is an ideal time to push forward with these new technologies and workflows. The Future Program Office in Action A Data User, such as a logistics of - ficer or cost estimator, needs to evalu - ate a proposed change to a radar sub - system. In the following phases, how would they access the correct data to properly assess program impact and make solid, data-informed decisions? • Phase 1. Document-Based: Wade through 500 pages of documents, manual searches, and days of work, only to find that the information may already be outdated. • Phase 2. MBSE with SysML v1: Dust off the SysML training from last year and wade through the model to hopefully find the data. Alternatively, email with/to the Modeler to request the artifacts and data that support your deci - sion, and wait. Then, you will have to ensure that you correctly inter - pret and understand the diagram and modeling notations. • Phase 3. SysML v2 + AI: Ask the AI agent for the data in natural lan - guage. The AI agent queries the federated digital thread and re - turns an accurate, traceable an - swer in common language. If you have questions or need additional information, just ask. Expanded across the program of - fice, this capability will enable acqui - sition professionals to find the cur - rent, authoritative data needed to make data-informed decisions that drive program success. Imagine a contracting officer providing a list of all proprietary-marked components in the system. Compare the results with the contract verbiage and pro - duce a list of any discrepancies that need to be addressed. Or imagine a program manager, when Component G58397-3 is at risk of Diminishing Manufacturing Sources and Material Shortages because a
Warfighting Acquisition University (WarU) uses a three- tiered framework (Figure 1) to describe digital engineering skills in the acquisition workforce. At the top sit Modelers, highly skilled experts who can build and manage models at all stages.
Cultural barrier. Even with the technology in place, we still face a cultural barrier, as model-based ac - quisition is a vast departure from our familiar legacy document-based approaches. Use of these tools goes against our traditional document- based acquisition legacy and pres - ents a completely new way of com - municating and developing systems. This huge shift has been recognized by the Software Engineering Institute , RAND Corporation , MIT, and others which have independently concluded that the main obstacles are organizational and cultural, not technical.
46 DEFENSE ACQUISITION MAGAZINE | SEPTEMBER –OCTOBER 2026
Made with FlippingBook - Online Brochure Maker