i watched the demo recordings from four hackathons last weekend. not because i'm a glutton for punishment. because someone has to.
here's what i found.
forty teams, one pitch deck
team one: "an AI agent that automates your inbox."
team two: "a smart assistant that manages your email workflow."
team three: "an intelligent layer between you and your messages."
team four: "GPT for email."
same product. four fonts. zero self-awareness.
and it wasn't just email. it was email, scheduling, meeting notes, customer support, and the inevitable "AI-powered to-do list that thinks for you." every category had five entries. every entry had the same architecture: a prompt, an API call, a dashboard nobody will open after sunday.
the demo that works exactly once
you know the moment. someone types "help me plan my day" and the agent responds with a beautiful structured plan. the crowd claps. the judges nod.
then someone types "what about next tuesday" and the whole thing dissolves into a hallucinated mess of conflicting calendar entries and a fake meeting with a person who doesn't exist.
the demo works once because it was built to work once. the prompt was tuned for that specific input. the architecture has no memory, no state, no recovery. it's a magic trick, not a product. and that's fine ... it's a hackathon ... but nobody's calling it that. they're calling it "the future of work."
the identity crisis, spelled out
here's where it gets uncomfortable.
when you ask a team "who is this for?" you get:
- "everyone who uses email"
- "busy professionals"
- "knowledge workers"
which means: nobody. you built a tool for a demographic, not a person. and that's why all forty projects feel identical ... because they're all aimed at the same ghost. the undefined user. the "anyone who needs productivity."
the identity crisis isn't the AI's. it's the builder's. you walked in with skills and a weekend and no idea what problem you actually care about. so you grabbed the nearest trending API and wrapped it in a UI that looks like every other UI because you used the same component library and the same color scheme and the same landing page template.
you didn't build a product. you built a mirror. and the mirror shows you what everyone else is building.
the one project that stuck
one team, at one hackathon, built something different.
it wasn't smarter. it wasn't faster. it was narrower. they built a tool for triage nurses at understaffed rural clinics to quickly categorize incoming patients by urgency using voice input. that's it. one workflow. one user type. one problem that exists in the real world right now.
the demo worked twice. then three times. then on a question they hadn't planned for, it stumbled ... but it stumbled in a way that was fixable, because the scope was small enough to see the edges.
they didn't win. the judges gave it to the email agent.
what to actually take from this
if you're building something this weekend, or next weekend, or the one after:
- pick one person. not a market. a person. someone you can name.
- solve one thing they do badly. not "improve their workflow." one thing.
- make the demo work twice. if it only works once, you don't have a product, you have a screenshot.
- stop calling it "AI-powered." if the AI is the most interesting thing about your product, your product isn't interesting.
hackathons aren't broken. builders aren't lazy. but when everyone reaches for the same hammer, everything looks like the same nail. and right now, the nail is "an AI agent that..."
be the weird one. be specific. be the team that builds for the triage nurse, not the "knowledge worker."
the internet has enough email agents. it doesn't have enough people who know exactly who they're building for.