Artificial intelligence has rapidly transformed software engineering, with generative AI and large language models (LLMs) now capable of producing vast quantities of code and documentation, and machine-learning algorithms adept at monitoring performance and detecting security vulnerabilities. However, a pivotal question has lingered: are these AI tools equally transformative when the task involves conceiving, designing, and fabricating a complex physical system, such as a modern jet engine? This past semester, the JARVIS Challenge (Jet-engine AI Research and Validation Intensive Sprint) at MIT set out to answer precisely this, tasking undergraduates with exploring AI’s potential to compress the notoriously lengthy and intricate design-build-test cycle, ultimately aiming to ascertain if AI could empower them to build faster and better.
The JARVIS Challenge emerged from a growing industry interest in leveraging advanced AI for applications beyond the digital realm, specifically in high-stakes hardware engineering where precision, safety, and physical understanding are non-negotiable. Traditional aerospace engineering cycles for complex systems like jet engines can span years, often decades, involving vast teams, extensive prototyping, and rigorous testing. The financial investment required can be astronomical, with development costs for a new commercial aircraft engine often exceeding several billion dollars. The prospect of significantly compressing this timeline, even by a fraction, holds immense implications for innovation speed, cost efficiency, and global competitiveness.
Professor Zolti Spakovszky, director of the MIT Gas Turbine Laboratory and a driving force behind the JARVIS Challenge, encapsulated the core finding: "The JARVIS challenge showed that AI can substantially accelerate safety-critical hardware engineering, but engineering judgment remains the decisive differentiator. 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. Manufacturing – not engineering design or analysis – remained the fundamental rate-limiting step." His insights underscore a nuanced relationship between human expertise and AI augmentation, suggesting a future where engineers don their role as conductors of AI, rather than mere users.
An Unprecedented Sprint: The JARVIS Challenge Parameters
The challenge provided an ambitious framework: undergraduates were given a mere four weeks to design, fabricate, assemble, and test a small gas turbine aero engine. Their primary engineering partner in this endeavor was AI. The objective was clearly defined: construct a "JARVIS-class" single-spool jet engine capable of producing between 50 and 100 pounds of thrust, running on Jet-A fuel, and completing five successful 60-second runs. Critically, teams were granted complete autonomy over design choices, material selection, and fabrication methods, fostering an environment of true experimental inquiry.
A diverse cohort of 31 students, representing nearly every department within the School of Engineering, organized into seven distinct teams. The teams themselves were varied, ranging from groups composed entirely of first-year students to those heavily weighted with seniors. This diversity meant that many participants arrived with limited or no prior experience in specialized fields such as turbomachinery, compressible flows, or even fundamental thermodynamics. For a significant number of the younger competitors, the JARVIS Challenge marked their very first encounter with the internal mechanics of a gas turbine engine, a testament to the challenge’s role as a potent, hands-on learning crucible.
Tools of the Trade: AI and Traditional Engineering Platforms
To support their intensive four-week sprint, the student teams had access to a comprehensive suite of resources. This included MIT’s state-of-the-art machine shops and established manufacturing vendors, providing the physical infrastructure necessary for rapid prototyping and fabrication. For design and analysis, commercial software packages such as Concepts NREC for turbomachinery design, SolidWorks for computer-aided design (CAD), and ABAQUS for finite element analysis (FEA) were at their disposal. Various test rigs were also provided for characterizing and assembling individual components, allowing for iterative refinement and validation.
Crucially, the teams were given unrestricted access to MIT Parley, a newly launched internal platform that aggregates frontier large language models through a single, streamlined interface. This privileged access allowed JARVIS faculty leads to directly observe how students interacted with the AI tools, providing invaluable data on prompt engineering, cost per prompt, specific LLM usage, and other critical interaction metrics. Early access to Parley was secured for all participants, and with substantial financial backing from MIT Lincoln Laboratory, the Department of Mechanical Engineering, and corporate sponsors including Safran, Voyager Technologies, Beehive Industries, and Boom Technology, students benefited from essentially unlimited AI usage.
The corporate sponsors were not merely benefactors; their involvement stemmed from a dual interest in recruiting emerging talent and a genuine curiosity about how AI might fundamentally reshape engineering workflows in the aerospace sector. Ryan (Hal) Hefron of Voyager Technologies underscored this sentiment, telling the students, "We see this as the future of engineering. You’re honing skills that are not just nice to have – they’re going to be the future baseline in the engineering workforce." Vincent Garnier, managing director of Safran Tech, echoed this enthusiasm, observing the competition with keen interest. "JARVIS was a genuine experiment, a learning endeavor. We frankly didn’t know what to expect, from the students or from the AI models," Garnier reflected. "What struck me coming from the students was: first, the enthusiasm to explore; then, as the project developed, they all came to the cool-headed realization of what AI could or could not help them with, and then almost instantly adapted for that. It makes me confident that this generation of leading engineers will probably not fall prey to easy and shortsighted use of AI, and will do so by keeping ever more in contact with experiments – physical or thought experiments."
Throughout the challenge, faculty leadership—comprising professors Zachary Cordero, Zolti Spakovszky, Masha Folk, and Andreea Bobu from the Department of Aeronautics and Astronautics, alongside engineers from Lincoln Laboratory and a dedicated team of teaching assistants—maintained a vigilant oversight, primarily to ensure safety. Weekly progress reviews served as critical checkpoints where student designs and their application of AI were rigorously evaluated. Professor Spakovszky, in particular, developed a subtle yet effective technique for guiding teams, often posing pointed questions like, "Do you know what a rabbet fit is? Take in the comment," nudging students toward crucial engineering considerations without directly providing solutions.
Navigating the Sprint: A Week-by-Week Chronology of Discovery
The four-week sprint unfolded as a microcosm of accelerated engineering development, revealing both the profound utility and the inherent limitations of AI in physical system design.
Week 1: Conceptualization and AI-Driven Exploration
The initial phase saw teams grappling with the fundamental principles of gas turbine design, a domain unfamiliar to many. AI quickly proved its worth as an invaluable knowledge accelerator. Teams leveraged LLMs like those available through Parley to rapidly summarize complex textbooks on turbomachinery, learn to operate sophisticated design software, identify potential vendors for specialized components, generate structured Excel sheets for data management, and conduct comparative analyses between various design architectures. One particularly innovative team even configured an AI agent within Parley to function as their dedicated project manager, demonstrating a creative application of the technology to organizational tasks. This early success solidified AI’s role as a potent research assistant and a facilitator of rapid knowledge acquisition, dramatically shortening the initial learning curve.
Week 2: Detailing Designs and Encountering AI’s Limits
As the challenge progressed into its second week, teams transitioned from conceptualization to the more granular tasks of detailed CAD design and the prototyping of combustors—a critical, high-temperature component of any jet engine. It was during this phase that the inherent limitations of current AI models began to surface. While LLMs such as Claude and ChatGPT were adept at suggesting design alternatives and bridging knowledge gaps, teams increasingly encountered issues such as "hallucinations" (AI generating plausible but factually incorrect information), "sycophancy" (AI agreeing with user prompts even when incorrect), and a fundamental lack of physical understanding. These shortcomings eroded student confidence and, paradoxically, began to slow down the design process as engineers spent valuable time verifying AI outputs. Elizabeth Tupaj, a member of team 811 Crew, articulated this sentiment clearly: "AI is a helpful tool, great at finding information, helping organize things, and can write well, but it can’t do design. The moment the engineer doesn’t know what is going on and the AI is in charge is the moment the design becomes unreliable, at least with AI at its present capabilities." Teaching assistant John Zhang noted the impact of these early frustrations: "Seeing this firsthand with the students reminded me how much first impressions matter. If the students couldn’t get answers from the AI early on, they quickly grew frustrated and formed a lasting opinion that precluded them from using it later."
Final Weeks: Fabrication, Assembly, and the Human Element
In the concluding weeks, as teams moved towards fabrication and assembly, another critical bottleneck emerged—one that no AI could solve: working with external vendors. Students reported significant difficulties in establishing rapport with new vendors identified through AI searches, many of whom showed little interest in accommodating the challenge’s extremely tight timelines. The teams ultimately found that the vendors who proved most responsive and reliable were those with whom the students or faculty mentors already had established personal relationships. This highlighted the enduring importance of human networks and established trust in the complex, real-world logistics of manufacturing, a domain where AI’s current capabilities fall short.
Moments of Triumph and Learning: Initial Test Results
Despite the hurdles, the pace of innovation was remarkable. Of the three finalist teams, "Fast and Fractured" achieved a significant early milestone: the first-attempt ignition of their mini-combustor. This team, despite having no prior gas turbine experience, had leveraged AI heavily for trade studies and architectural comparisons, ultimately arriving at a viable design. The successful ignition was a powerful affirmation of the challenge’s premise.
Professor Masha Folk, the Charles Stark Draper Career Development Professor of Aeronautics and Astronautics, vividly recalled the moment: "The JARVIS Challenge showed what’s possible when you combine AI-enabled design with motivated students and a culture of rapid experimentation. The moment that stood out most was when the first student-designed combustor was installed on the test stand. It ignited flawlessly, ramped to full power, transitioned to dual-fuel operation, and then sustained stable combustion on 100 percent Jet-A fuel. This was proof that we can dramatically accelerate the cycle of design, build, and test while giving students hands-on experience with a real engineering challenge."
The Ultimate Test: Full Engine Trials and Final Outcomes
By the end of May, the two more senior teams, "Fast and Fractured" and "811 Crew," had progressed to full engine tests. Fast and Fractured, whose design was heavily AI-assisted, faced persistent delays due to vendor issues throughout the competition. When their engine finally made it to the test stand for a hot fire, the run was unfortunately cut short as the rotor rubbed and seized against the stationary housing, a common failure mode in turbomachinery.
However, Team 811 Crew, which entered the competition with more prior exposure to turbomachinery and propulsion concepts, emerged victorious. Their meticulously designed engine started successfully, transitioned seamlessly to Jet-A fuel, and generated measurable net thrust. PhD student Joe Chiapperi, reflecting on the intensity of these moments, described the palpable tension: "As we stood there with the air-starter, hearing their engines spool up and watching them spit fire, it felt like my heart was racing out of my chest. There were so many ways it could go wrong! What these students accomplished in such a short time span is nothing short of amazing."
Interestingly, the 811 team had displayed a marked resistance to heavy AI reliance throughout the competition, instead prioritizing their foundational knowledge, engineering fundamentals, and collaborative teamwork. Elizabeth Tupaj elaborated on their strategy: "We had people who were at least somewhat familiar with the design software, mechanical engineers who knew how to build anything, and aerospace engineers who had taken classes on the design of gas turbine engines specifically." This combination of specialized human expertise proved decisive.
The Human Differentiator: Experience, Judgment, and AI-Native Engineering
The JARVIS Challenge presented a compelling dichotomy: younger students, less encumbered by traditional methodologies, tended to use Parley more frequently and creatively from the outset. In contrast, the juniors and seniors often leveraged their deeper, pre-existing engineering experience. This observation led to a crucial insight regarding the optimal integration of AI.
Professor Andreea Bobu articulated this "sweet spot" of AI adoption: "JARVIS taught me that getting value from AI takes two things: enough expertise to judge what it tells you and catch it when it’s wrong, and enough curiosity to actually lean on it where it could help. The team that moved fastest in the sprint was experienced and leaned heavily on AI to get there. The team that eventually won was more resistant to AI; they had the expertise, but that skepticism made them slower. 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. To me, that’s the real opportunity ahead: training the next generation of engineers who have the judgment to direct these AI tools and the instinct to reach for them."
The competition’s clearest and most profound finding was that engineering experience acts as a powerful multiplier, and the human factor remains an absolutely vital element in complex system design. Mastery of first principles and fundamental concepts cultivates sound engineering judgment—the ability to navigate complex decision trees, assess risks, and make critical choices even in the face of incomplete or ambiguous information. When it comes to designing and building safety-critical physical systems, the irreplaceable elements are human hands, human intuition, and ultimately, human accountability. Teaching assistant Kyle Woody summarized this succinctly: "JARVIS has shown that AI copilots can have a multiplicative effect on engineering productivity, with judgment and first-principles thinking serving as the key differentiators among teams." Professor Cordero, associate director of the MIT Gas Turbine Laboratory, further emphasized the role of education: "Performance in JARVIS correlated strongly with year in school. My main takeaway is that in the AI era, education is more valuable than ever."
Broader Implications: Reshaping Aerospace R&D and Workforce Dynamics
The implications of the JARVIS Challenge for the aerospace industry and beyond are significant and far-reaching. If small, agile teams, effectively utilizing well-managed AI copilots, can genuinely compress design-build-test cycles from years down to mere weeks, the ramifications for global R&D timelines, cost efficiency, and competitive dynamics are profound. This acceleration could democratize access to complex hardware development, enabling smaller startups to compete with established giants, and potentially usher in an era of more rapid prototyping and iterative design in fields traditionally characterized by long lead times and high barriers to entry.
The challenge highlights an impending shift in workforce structure, accelerating the demand for what Professor Spakovszky termed "AI-native engineers." These are not just engineers who use AI, but those who understand its capabilities and limitations deeply enough to lead it, orchestrating its application to solve complex problems while retaining critical human oversight. This will necessitate a re-evaluation of engineering education, placing even greater emphasis on foundational principles, critical thinking, and hands-on experience, ensuring that future engineers possess the judgment to direct powerful AI tools.
Ultimately, the JARVIS Challenge served as a real-world crucible where students grappled with the stakes of AI integration not as a theoretical exercise, but within the tangible confines of a machine shop, with a live jet engine on a test stand. It underscored that while AI offers unprecedented tools for acceleration and optimization in hardware engineering, the ingenuity, critical judgment, and accountability of the human engineer remain the bedrock of innovation and safety in the design of physical systems.
