Blab Tech
MIT JARVIS Challenge Shows AI Can Speed Design of Small Jet Engines, but Human Judgment Remains Key
In a whirlwind four‑week sprint held at MIT this spring, undergraduate teams raced to design, build, and test a single‑spool gas‑turbine engine that could produce 50–100 pounds of thrust on Jet‑A fuel. The exercise, dubbed the JARVIS Challenge (Jet‑engine AI Research and Validation Intensive Sprint), was a laboratory for testing whether large language models (LLMs) and other generative AI tools could compress the traditionally long design‑build‑test cycle for a safety‑critical physical system.
The competition drew 31 students from MIT’s School of Engineering, split into seven teams. Participants ranged from first‑year freshmen to seniors, many of whom had never studied turbomachinery or compressible flow before the event. They were granted full access to MIT’s machine shops, commercial CAD and simulation software, and a newly launched platform called Parley that aggregates frontier LLMs through a single interface. Parley also logged the prompts, costs, and model types used by each team.
According to the challenge organizers, the teams could choose any design, materials, or fabrication method. The only constraints were that the engine must produce the required thrust, run on Jet‑A, and complete five 60‑second test runs. Sponsors—including Safran, Voyager Technologies, and Beehive Industries—provided financial support that enabled unlimited AI usage for the students.
In the early days of the sprint, AI proved a valuable aide for knowledge acquisition and routine tasks. Teams used LLMs to summarize textbooks, learn how to use design software, source vendors, create spreadsheets, and perform comparative analyses of design decisions. One team even built an AI agent in Parley to act as a project manager.
However, by the second week the limitations of generative AI became apparent. While Claude and ChatGPT could suggest design alternatives, the models frequently produced hallucinations or lacked physical understanding. A member of team 811 Crew noted that “AI is a helpful tool, great at finding information, helping organize things, and can write well, but it can’t do design.” The team’s experience highlighted that when engineers lack confidence in the AI’s output, they may abandon the tool altogether.
Vendor selection also proved a hard problem for AI. Teams found that AI‑identified suppliers had no interest in meeting the tight timelines, and only vendors with whom the students had pre‑existing relationships could deliver parts quickly.
Despite these challenges, the competition produced tangible results. Two teams—Fast and Fractured and 811 Crew—completed full engine tests by the end of May. Fast and Fractured, which relied heavily on AI for trade studies and architecture comparisons, achieved first‑attempt ignition of their combustor but suffered a rotor seizure during hot‑fire testing. 811 Crew, which had more experience with turbomachinery and was initially skeptical of AI, successfully started their engine, transitioned to Jet‑A, and produced net thrust.
Professor Zolti Spakovszky, director of the MIT Gas Turbine Laboratory, said the event demonstrated that AI can accelerate safety‑critical hardware engineering, but that engineering judgment remains the decisive factor. “An AI‑native engineer is not defined by using AI, but by leading it—knowing when to trust it, when to challenge it, and how to translate AI outputs into working hardware,” he explained.
The competition’s findings suggest that while AI can reduce the time spent on routine design tasks, the manufacturing step remains the fundamental bottleneck. The human factor—expertise, skepticism, and the ability to catch AI hallucinations—continues to be essential for building reliable physical systems.
Looking ahead, the JARVIS Challenge indicates that small teams equipped with well‑managed AI copilots could compress design‑build‑test cycles from years to weeks. This shift could reshape workforce structures, research and development timelines, and competitive dynamics in aerospace and other high‑stakes engineering fields. The students who tackled the challenge are among the first engineers to experience these possibilities in a real machine‑shop setting.
The event also underscored the importance of education. Professor Andreea Bobu noted that the team that moved fastest was experienced and leaned heavily on AI, while the winning team was more resistant to AI but had the expertise to stay in charge of the tool. “The sweet spot seems to be knowing enough to stay in charge of the tool, and being eager enough to pick it up in the first place,” she said.
In sum, the MIT JARVIS Challenge confirms that AI can multiply engineering productivity, but that judgment and first‑principles thinking remain the key differentiators. The next generation of engineers will need both the technical knowledge to evaluate AI outputs and the curiosity to use AI where it can help.
The competition drew 31 students from MIT’s School of Engineering, split into seven teams. Participants ranged from first‑year freshmen to seniors, many of whom had never studied turbomachinery or compressible flow before the event. They were granted full access to MIT’s machine shops, commercial CAD and simulation software, and a newly launched platform called Parley that aggregates frontier LLMs through a single interface. Parley also logged the prompts, costs, and model types used by each team.
According to the challenge organizers, the teams could choose any design, materials, or fabrication method. The only constraints were that the engine must produce the required thrust, run on Jet‑A, and complete five 60‑second test runs. Sponsors—including Safran, Voyager Technologies, and Beehive Industries—provided financial support that enabled unlimited AI usage for the students.
In the early days of the sprint, AI proved a valuable aide for knowledge acquisition and routine tasks. Teams used LLMs to summarize textbooks, learn how to use design software, source vendors, create spreadsheets, and perform comparative analyses of design decisions. One team even built an AI agent in Parley to act as a project manager.
However, by the second week the limitations of generative AI became apparent. While Claude and ChatGPT could suggest design alternatives, the models frequently produced hallucinations or lacked physical understanding. A member of team 811 Crew noted that “AI is a helpful tool, great at finding information, helping organize things, and can write well, but it can’t do design.” The team’s experience highlighted that when engineers lack confidence in the AI’s output, they may abandon the tool altogether.
Vendor selection also proved a hard problem for AI. Teams found that AI‑identified suppliers had no interest in meeting the tight timelines, and only vendors with whom the students had pre‑existing relationships could deliver parts quickly.
Despite these challenges, the competition produced tangible results. Two teams—Fast and Fractured and 811 Crew—completed full engine tests by the end of May. Fast and Fractured, which relied heavily on AI for trade studies and architecture comparisons, achieved first‑attempt ignition of their combustor but suffered a rotor seizure during hot‑fire testing. 811 Crew, which had more experience with turbomachinery and was initially skeptical of AI, successfully started their engine, transitioned to Jet‑A, and produced net thrust.
Professor Zolti Spakovszky, director of the MIT Gas Turbine Laboratory, said the event demonstrated that AI can accelerate safety‑critical hardware engineering, but that engineering judgment remains the decisive factor. “An AI‑native engineer is not defined by using AI, but by leading it—knowing when to trust it, when to challenge it, and how to translate AI outputs into working hardware,” he explained.
The competition’s findings suggest that while AI can reduce the time spent on routine design tasks, the manufacturing step remains the fundamental bottleneck. The human factor—expertise, skepticism, and the ability to catch AI hallucinations—continues to be essential for building reliable physical systems.
Looking ahead, the JARVIS Challenge indicates that small teams equipped with well‑managed AI copilots could compress design‑build‑test cycles from years to weeks. This shift could reshape workforce structures, research and development timelines, and competitive dynamics in aerospace and other high‑stakes engineering fields. The students who tackled the challenge are among the first engineers to experience these possibilities in a real machine‑shop setting.
The event also underscored the importance of education. Professor Andreea Bobu noted that the team that moved fastest was experienced and leaned heavily on AI, while the winning team was more resistant to AI but had the expertise to stay in charge of the tool. “The sweet spot seems to be knowing enough to stay in charge of the tool, and being eager enough to pick it up in the first place,” she said.
In sum, the MIT JARVIS Challenge confirms that AI can multiply engineering productivity, but that judgment and first‑principles thinking remain the key differentiators. The next generation of engineers will need both the technical knowledge to evaluate AI outputs and the curiosity to use AI where it can help.
