Air Force Cadet Built Military AI App With No Coding Experience

A U.S. Air Force cadet with zero programming background successfully built a functional military software application using only AI chatbots — a proof-of-concept that could reshape how service members develop tactical tools.

Joshua Lynch, working through the U.S. Department of the Air Force–MIT AI Accelerator's Phantom Program, spent three months using ChatGPT, Claude, and Google Gemini to create what he called the Remote Operating Modular Augmentation Device (ROMAD-AI). His mentor, Laura Niss from MIT Lincoln Laboratory's Embedded and AI Systems Group, tracked his progress to understand whether "vibe-coding" — relying entirely on prompts to guide AI chatbots through code generation — could work for military applications.

Lynch initially envisioned an ambitious battlefield assistant with AI-powered target recognition, autonomous capabilities, and communication management. Reality proved more constrained. By project's end, he had pivoted to a document-processing tool that analyzes tactical maps and generates mission-planning documents through a chatbot interface.

Why it matters

Military software development typically requires specialized programmers and moves through lengthy procurement pipelines. If service members with domain expertise but no coding skills can prototype applications themselves, they could accelerate innovation cycles and better translate operational needs into technical requirements. The research suggests AI chatbots work well for rapid prototyping, though production deployment of sensitive military systems still requires expert code review and security validation.

Learning the limits

Most of Lynch's timeline was consumed learning to work around chatbot limitations rather than building features. The AI systems frequently modified unrelated code sections and lost focus on the task hierarchy. Lynch discovered he needed to break problems into small components, frame questions precisely, and constantly redirect conversations back to objectives.

One security incident highlighted the risks: Lynch didn't initially realize his final application was sending input documents to Gemini's cloud servers for analysis rather than processing them locally — a critical vulnerability for military data.

Niss tracked how Lynch's perception of AI capabilities evolved. He started with outsized expectations and ended with a grounded understanding of current technology limits. Among the three chatbots, Claude showed more stability than ChatGPT across measures including perceived intelligence and consistency.

From prototype to production

The final application, built using Google AI Studio App, demonstrated viability but fell short of deployment readiness. It lacked the security controls required for battlefield use and didn't deliver all originally planned capabilities.

Niss came away impressed by the prototype quality from a nontechnical user. "I'm now of the opinion that these can be powerful tools for nontechnical experts to convey problems and possible solutions to technical experts, and aid in communicating desired outcomes," she said.

The research reinforces that AI chatbots function better as prototyping assistants than production tools for critical applications. Code review remains a bottleneck — AI can generate substantial functional code, but human experts must validate security and correctness.

"No matter how good AI gets, I think we'll always need to collaborate to get to the best solutions for the most important problems," Niss noted.

The research was sponsored by the Department of the Air Force Artificial Intelligence Accelerator under Cooperative Agreement Number FA8750-19-2-1000. Details were first reported by MIT Lincoln Laboratory.