The post says software hackathons no longer test software skill in any meaningful way because AI can spit out passable prototypes fast enough that the winner is often whoever tells the best story, not whoever hacked the hardest. It argues hardware still resists this because physical constraints are harder to fake, so building with electronics, sensors, or custom enclosures keeps the event grounded in actual making.
That basic diagnosis landed with a lot of people, but the sharper point was that this rot is older than LLMs. Plenty of people said software hackathons had already become demo theater years ago. Winning entries were often polished landing pages, mock data, bought
Bootstrap themes, or even straight PowerPoints. Corporate versions got singled out as the worst form. They reward presentation, let teams show work started weeks earlier, and sometimes turn weekend experiments into uncompensated product work for management.
The most useful distinction people made was not "software versus hardware" but "competition versus collaboration." Older hackathons were described less as contests and more as short, intense build sprints where the point was to learn, ship something rough, and push yourself. Once judging, prizes, sponsor agendas, and
VC-style pitching took over, the event started optimizing for optics. That is why several people were less interested in banning AI than in changing incentives. They want code review, stricter limits on prebuilt work, technically literate judges, or no winners at all.
Hardware still got genuine enthusiasm because it produces something tangible that is easier to explain and harder to fake. People pointed to
Arduino and
Raspberry Pi builds,
PCB work, Shenzhen-style fast-turn prototyping, and student programs like Hack Club as proof that physical projects can be thrilling and accessible. But even here, commenters warned against romanticizing it. Much of what gets called a hardware hackathon is still mostly software on top of commodity boards, and AI can already help with design, iteration, and instrumentation. The case for hardware was less that it is AI-proof and more that physical reality imposes a tighter feedback loop than prompt-driven demos do.