U.S. Air Force cadet Joshua Lynch, under the mentorship of MIT Lincoln Laboratory’s Laura Niss, successfully developed a military-focused AI application using major chatbots from Anthropic, OpenAI, and Google. His effort, part of the Department of the Air Force–MIT AI Accelerator’s Phantom Program, illustrated how beginners with no prior coding experience can harness generative AI tools to create defense-relevant software prototypes.
What Happened
Over three months, Lynch employed a technique dubbed “vibe-coding,” where he guided generative AI chatbots solely through textual prompts to write and refine code. His primary tools were the paid versions of Anthropic’s Claude, OpenAI’s ChatGPT, and Google’s Gemini models. The project aimed to develop a modular military application for battlefield support, encompassing features like AI-assisted target recognition, intelligence modularity, autonomous striking, and communication management.
Initially envisioned as a comprehensive tool to reduce collateral damage and improve mission survivability, Lynch had to scale back the application due to technical and security limitations inherent in the chatbot systems. The resulting prototype, named Remote Operating Modular Augmentation Device (ROMAD-AI), focused on basic document processing tasks such as tactical map analysis and mission planning document generation. The final build utilized Google AI Studio App to interface with the Gemini API and integrate AI directly within the development environment.
Key Facts
The research was sponsored by the Department of the Air Force Artificial Intelligence Accelerator under Cooperative Agreement Number FA8750-19-2-1000. Joshua Lynch was mentored by Laura Niss from MIT Lincoln Laboratory’s Embedded and AI Systems Group. The AI chatbot products used included Anthropic’s Claude, OpenAI’s ChatGPT, and Google’s Gemini, all accessed primarily through their web-based chat functions, except for the final product integration via Google AI Studio App.
The completed prototype does not yet meet operational security standards for battlefield use and lacks full functionality originally planned. Lynch encountered challenges such as the AI’s lack of hierarchical coding focus and the need to break problems down for effective prompt management. The project highlighted that while AI can generate substantial code, human review remains essential to mitigate security risks.
What This Means
Lynch’s project demonstrates a practical shift in military software development, showing that nontechnical personnel with domain knowledge can directly engage with AI chatbot tools to prototype solutions. This potentially shortens traditionally lengthy and costly development pipelines by enabling concept validation and early design iterations without extensive programming expertise.
Such democratization of code generation could empower service members across ranks to contribute innovations tailored to tactical needs, increasing responsiveness to emerging challenges in warfare. However, the project also underscores the risks of premature deployment without rigorous code vetting, especially in sensitive defense environments.
The research underscores that, despite advances in AI coding, collaboration between domain experts and technical specialists remains critical to produce secure, effective software. It also points to the evolving role of major tech companies’ AI products in defense innovation, providing foundational tools that can be adapted beyond traditional development frameworks.
Background
The Phantom Program, under the U.S. Department of the Air Force and MIT AI Accelerator collaboration, promotes the integration of AI technology to solve military problems. Lynch’s project aligns with this effort by exploring how generative AI tools—often designed for commercial uses like ChatGPT and Claude—can be adapted for military applications by nonexperts.
Prior to this project, AI chatbots were mainly studied as aids for tasks like email composition and data analysis. This work represents one of the first systematic attempts to leverage these models for tactical military software development driven by a coding novice.
What Remains Unclear
The security implications of sending sensitive battlefield data to cloud-hosted AI models like Gemini remain unresolved. The full extent of code vulnerabilities generated by AI and how best to mitigate them through human review is still under evaluation. Additionally, the scalability of AI-assisted vibe-coding for broader military software development beyond prototypes is yet to be determined.
What Comes Next
The Department of the Air Force Artificial Intelligence Accelerator may pursue further studies evaluating AI-enabled prototyping with other nontechnical users. There is no confirmed timeline for transitioning prototypes like ROMAD-AI into operational settings, pending resolution of security and functional limitations.
Sources
This article is based on reporting and publicly available information from the following sources:
Read more Artificial Intelligence stories on Goka World News.