Last spring I judged two hackathons three weeks apart. The first was a sprawling general event with 400 people and a prize sheet that read like a department store catalog: best fintech, best healthtech, best use of some sponsor's API, best social good. By Sunday afternoon I'd seen the same to-do app reskinned eleven times. The second was a 60-person climate hackathon focused entirely on grid optimization and carbon accounting. Every single project was something I'd want to follow up on. One team had a working prototype for matching rooftop solar surplus to nearby EV chargers. I still have their deck.

That contrast has stuck with me, and it's not an isolated data point. After roughly a decade of attending these things, I've become convinced that the general-purpose hackathon is quietly dying, and good riddance. The future belongs to the narrow ones.

The tyranny of the blank canvas

Here's the thing nobody tells first-time organizers: constraints are a gift. When you tell 400 strangers "build anything," you've handed them a blank canvas, and a blank canvas is paralyzing. People default to what they already know how to build. That's why you get the eleventh to-do app. The cognitive overhead of choosing a problem eats most of the weekend, and teams spend Saturday morning arguing about what to build instead of building.

A domain hackathon removes that tax. When the theme is "reduce friction in clinical trial recruitment" or "tools for open-source maintainers," the problem space is already mapped. Teams arrive having read the brief, sometimes having half-formed an idea on the train. I've watched the energy in the room at a focused event versus a general one, and it's not close. At the climate event, people were soldering and arguing about kWh by 10am. At general events, that hour is spent on Post-it notes.

Niche events attract people who actually know things

This is the part organizers underrate. A general hackathon pulls a random cross-section: students padding resumes, recruiters lurking, a few genuine builders. A domain hackathon self-selects for people who care about that domain enough to give up a weekend for it.

At an AI hackathon focused specifically on retrieval and agents, the person next to you is likely someone who has shipped a RAG pipeline to production and has opinions about chunking strategies. At a climate hackathon, you'll find a grid engineer, a carbon-markets analyst, and a satellite-imagery person in the same room. The conversations are denser. The feedback is sharper. You learn more in two days than you would in two months of reading.

There's a network effect here too, and it compounds. Roughly 70% of the people I now consider genuine peers in the climate-tech space, I met through three or four niche events, not the big general ones. The signal-to-noise ratio of who you meet is just dramatically higher when the room is filtered by genuine interest.

The counterintuitive part: niche events produce more reusable work

You'd think a narrow theme would produce narrow, throwaway projects. The opposite is true, and this is the point I'd push hardest.

Because everyone is solving adjacent problems, the building blocks teams create tend to be useful to each other. At that grid-optimization event, three different teams needed the same utility-rate dataset, so someone cleaned it and shared it by Saturday night. At general hackathons, every project is an island; nothing connects. The domain focus creates a shared substrate of tools, data, and conventions that outlives the weekend. I've seen open-source libraries born at niche hackathons that are still maintained two years later. I cannot name a single durable artifact from a general one.

This is also why sponsors increasingly prefer them. A company selling developer infrastructure gets far more from 60 engineers who deeply understand the problem than 400 who'll forget the brand by Monday.

How to actually pick one

If I've convinced you, here's the practical part. Don't optimize for prize money or headcount. Optimize for fit. A few things I look for:

  • A specific, opinionated theme. "AI for X" is still too broad. "Evaluation tooling for LLM agents" is the sweet spot.
  • Organizers who are practitioners. If the people running it work in the domain, the brief and the judging will be sharper.
  • A community that persists afterward. Is there a Discord, a follow-up meetup, a recurring event? The weekend matters less than what comes after.
  • Realistic scope. The best briefs are narrow enough to ship something real in 48 hours.

Finding these used to be the hard part, because niche events don't have the marketing budget of the big sponsored ones. That's gotten easier. I use Droppa to filter by domain and city, which is genuinely how I found the grid-optimization event I keep going on about. If you want to find events on Droppa that match a specific interest rather than scrolling through generic listings, it saves a lot of time.

My one piece of advice: pick the narrowest event that still excites you, and go alone if you have to. The constraint is the feature. You'll build something real, meet people who actually know the space, and walk away with more than a t-shirt. That's a weekend well spent.