In today’s world, artificial intelligence chatbots like ChatGPT and Claude can perform many functions, such as writing work emails and planning travel routes. These chatbots are systems based on large vision-language models (VLMs): AI trained on a huge data set that includes books, websites, code and images.
The AI algorithms are then refined using massive amounts of human-generated feedback to follow instructions and avoid harmful or unwanted output, and use this “knowledge” to create text or images based on a user’s input. Although chatbots have clear limitations, they can be very useful for a variety of tasks, including some areas that traditionally require specialized skills, such as computer programming.
As part of a project for the U.S. Department of the Air Force – MIT AI Accelerator’s Phantom Program – U.S. Air Force cadet Joshua Lynch wanted to see – with the help of his mentor Laura Niss, a technical associate in the Embedded and AI Systems Group at MIT Lincoln Laboratory – whether he could develop a fully functional program as a complete novice to programming. He used a process called “vibe coding,” in which a user relies entirely on prompts to guide a generative AI chatbot in writing and refining code.
His motivation was to enable anyone familiar with the military problem space, regardless of their technical background, to advance their ideas for useful software applications, essentially bypassing the time and cost constraints of the traditional military software development pipeline. Lynch wanted to develop his own application while Niss oversaw his experience with the technology.
“The Phantom student wanted to see if he could create a useful application through self-identified vibe coding without prior knowledge,” says Niss. “As part of this project, I wanted to understand how his perception of AI changed over time with use. We both wanted to better understand where and how AI can be used by non-technical users in the military.”
Lynch set out to see if he could create an application specifically for his type of tactical team with no programming knowledge and using chatbots to reduce collateral damage while improving survivability in the broader mission. This application would offer features including AI-powered target detection; modular intelligence, surveillance and reconnaissance; autonomous beating; and battlefield communications management.
During the project, Lynch completed several AI training courses and became familiar with the technology’s military and non-military uses. As a basis for his code generation, he used the paid models of three AI chatbots: Claude from Anthropic, ChatGPT from OpenAI and Gemini from Google. Most of this work was done only through the chatbots’ main chat function in a web browser, not as an integrated system within a development environment, as is standard today. The final application was built using the Google AI Studio app, which can create applications that interact with the Gemini application programming interface and are integrated into the development environment via AI.
For three months, Lynch worked with these models to develop his application called Remote Operating Modular Augmentation Device (ROMAD-AI). During this time, he learned various methods to improve code output. For example, he often encountered difficulties because the AI chatbots lacked hierarchical focus and modified unrelated sections of code. He discovered that it is important to break problems down into small parts, formulate questions clearly, and bring conversations back on topic when they stray too far from the goal.
Most of the project time was spent learning the limitations of chatbots and how to work around them effectively. As Lynch gained more experience with the chatbots, limitations in AI capabilities and development time led him to change the scope of the project, moving it from an application that could assist on the battlefield to one that could perform basic document processing, such as analyzing tactical maps of battlefields and creating mission planning documents via an interface with a VLM-powered chatbot. Although the resulting prototype did not fulfill all of the functions that Lynch had originally intended (and in its current version was not safe for the desired use case), it demonstrated the power and usefulness of such an application for military personnel.
“I was quite impressed with this final product and it showed me how powerful these systems can be in prototyping non-professional designs,” says Niss. “I now believe these can be powerful tools for non-technical experts to communicate problems and possible solutions to technical experts and to help communicate desired outcomes.”
Niss observed the change in Lynch’s view of AI language models during his experience. After starting with an impressive goal, Lynch gained an understanding of the capabilities of current technology and significantly narrowed his expectations by the end of the project period. Measurements of his perception of the various AI systems over time and across system updates were of particular interest to Lynch and Niss, with Claude showing greater stability than ChatGPT on traits such as likeability, anthropomorphism, and perceived intelligence. Lynch found the AI to be a helpful tutor, but noticed that it had inaccuracies on topics he knew well.
The project showed that AI chatbots can empower non-technical service professionals to create usable software applications for their unique problems, although they work better as a prototyping assistant than as a full production tool when handling sensitive information and for critical applications. Improper code review can lead to security risks, as demonstrated by a case in which Lynch failed to recognize that the final application was sending the input documents to a Gemini AI model for analysis, rather than analyzing the documents locally on his computer. Although AI can generate significant amounts of functional code, code review remains a bottleneck in this area.
“For me, this project strengthened collaboration between experts in different fields,” says Niss. “No matter how good AI gets, I think we always need to work together to find the best solutions to the most important problems.”
The research was funded by the Department of the Air Force Artificial Intelligence Accelerator and conducted under Cooperative Agreement No. FA8750-19-2-1000.
https://news.mit.edu/2026/how-novice-coders-can-develop-ai-programs-for-military-applications-0707
