week to complete were now accom- plished in just one or two hours by a single person. Moreover, the quality of the team’s planning outputs improved dramatically. The S.M.A.R.T. goals set for the team became more cohesive, comprehensive, and clearly defined and reflected deeper strategic intent. This newfound efficiency and en- hanced quality were not just incre- mental improvements; they repre- sented a paradigm shift in how the team approached work. It was clear that AI, wielded effectively, could be a game-changer for SSC. However, this initial success was just the be- ginning. Integrating AI into a large organization like SSC presented its own unique challenges. Navigating AI Adoption Integrating this new technology into established workflows and or- ganizational cultures was a complex undertaking. The following key chal- lenges highlight the complexities of widespread AI implementation. The learning curve. Mastering NIPRGPT, like any new technology, required an investment of time and ef- fort. Fortunately, the Warfighting Ac- quisition University offered resources such as free online courses taught by leading AI experts on platforms like Coursera. These courses provided a foundation in AI concepts and tech- niques, including prompt engineer- ing and risks and biases. Initially, the focus on using LLMs was for relatively simple tasks, such as refining written work and generating ideas—treating AI more as a sounding board or re- viewer than a content creation tool. The biggest initial hurdle was simply knowing where to start. Ironically, this was overcome by asking NIPRGPT it- self how it could best assist, which proved surprisingly effective in iden- tifying potential use cases and maxi- mizing its utility. Addressing awareness and ac- cess. Unlike with other emerging technologies, skepticism about NI- PRGPT wasn’t the primary barrier to adoption. Instead, the biggest chal-
complex tool for which there never seemed enough time to learn. That all changed during a post- Thanksgiving lull. In standing up a new USSF program, the current task at hand was to manually create countless Jira (planning tool) tickets to translate the team’s strategic road- map into actionable steps. The sheer volume of work was overwhelming. And in a moment of near despera- tion, NIPRGPT was used to explore untapped potential. Feeding the program’s strategic plan into the system involved using prompt engineering to generate Jira tickets complete with descriptions, definitions of done, and acceptance criteria. Essentially, the system was asked to transform the strategic vision into S.M.A.R.T. (Specific, Measurable, Actionable, Relevant, Time bound) goals. The results were a revelation (a tailorable prompt is included with this article). The output that materialized was like getting a glimpse into The Matrix , a 1999 movie where everyone has the individual responsibility to make the choice between the real world and an artificial world. A sudden understand- ing of the AI’s vast untapped potential became impossible to discount. The
implications are vast, as making this capability available to all Guardians will revolutionize USSF workflows and processes. But with the opportunities come risks. Multiple dangers must be navigated: These include halluci- nations (where AI “makes up facts”) and biases (toward agreeing with the one using AI or those who trained it). It was evident that AI literacy courses are essential requirements for proper AI management and implementa- tion when the larger workforce starts adopting it. The following two weeks became a whirlwind of exploration and experi- mentation. Focusing on integrating NIPRGPT into existing workflows, this LLM was able to generate not only Jira tickets but also code for automating various tasks. There were hurdles as the systems used weren’t designed for seamless AI integration, requiring a “human in the loop” to manually debug and re- fine LLM-generated code using tools like WordPad and Jira. Despite the manual effort involved, this hands- on experience provided invaluable insights into how LLMs could enhance and automate even the most complex processes. Tasks that previously re- quired a team of two or three a full
20 | DEFENSE ACQUISITION | November-December 2025
Made with FlippingBook - Online Brochure Maker