Finishing a hackathon project used to mean something. Now it just means you had a good prompt.
Someone on my timeline said the golden age of hackathons is dead. 2021 to 2023, they said, was the easiest way to build a network, learn fast, and land a job without touching Leetcode. Then AI nuked it. Building got so easy that the "idea guys" took over, and nobody's sure what replaces hackathons for someone starting out today.
I don't fully agree. But I get exactly what they're mourning.
Back before AI tools were part of my workflow, I entered a few hackathons, some Google ones among them, and honestly my memory of most of it is pretty vague now. What I do remember clearly is not finishing. Not because the idea was bad. I was stuck in tutorial hell, dealing with imposter syndrome on top of it.
I still call those hackathons rewarding. Not because I shipped something, most of the time I didn't, but because they showed me what was actually obtainable in this industry. Real working software, built by people my age, solving real problems in a weekend. Before that exposure, "good" was an abstraction. After it, "good" had a shape.
Right now I'm juggling a handful of hackathons at once. Agents for Humans, AI Builders, the Niger Delta VES fleet build, a few others. The biggest difference from those early days isn't the tech. It's that I actually finish now.
I get bored of most games after an hour. Building doesn't do that to me. Endless progress. Rewarding. Mine when it's done. People don't always get that building is the relaxing thing, not the job.
Which is exactly why finishing matters more now, not less. If building is the game, unfinished hackathons are just levels I never beat. At this point somebody needs to electrocute me if I so much as glance at a new project before these are done.
AI didn't fix tutorial hell. It just made it optional. The blocker isn't syntax anymore, or four hours lost debugging an import error at 2am. That friction is mostly gone.
But AI enhances more than speed. It enhances your ability to build the wrong thing, confidently, on schedule, with a working demo to show for it.
That's basically what the "idea guys taking over" complaint is pointing at. Finishing used to be proof you understood what you built, because finishing was hard. Now finishing is available to anyone with a good prompt and no real judgment about whether the thing being built is worth building.
Tutorial hell was miserable. I'm not nostalgic for it. But debugging your own broken import statement at 2am taught you what actually breaks and why. A working AI-generated demo doesn't.
I don't think the answer is bringing that friction back on purpose. I don't think the industry has figured out what replaces it either. AI removed the thing that used to force people to learn slowly. Nobody's built the thing that forces people to learn fast without skipping the part where you actually understand what you shipped.
Hackathons still do the thing that mattered most for me. They show you what's obtainable, who's building it, what the bar actually looks like up close. That part isn't dead. I'm getting it right now, across every one of these I'm juggling.
What's dead is the assumption that finishing proves anything. Finishing used to mean something. Now it just means you had a good prompt.
Top comments (8)
There have always been at least two kinds of hackathon participants in my experience. Those taking the hacking spirit seriously, collaborating and innovating, and then the ones eager to win and finish, which often implied something very close to cheating, like coming overly prepared and reusing prior work, focusing on what judges liked to see.
AI assistance and agentic vibe coding to me feels like playing in a lottery. Some rare times, it generates working code quickly and helps me understand things, in most cases it's still a waste of time, and "tutorial hell" would have been much better, if it wasn't for the search result enshittification making it hard to find relevant sources, the neglected and outdated documentation, and the marketing copywriting flooding tutorials with promotion and slop.
I don't need to bring back friction on purpose. There will always be fricition eventually.
And it still is. Yes, with vibe coding agent monkeys rolling the dice until no tests fail, the "idea guys" can produce working prototypes good enough to impress other idea gues with a hackathon demo. But as we saw, they could do that before AI, they just had to team up with an eager developer back then. All of that is still far from maintainable and secure production-grade software though.
Shipping software and maintaining and supporting it in production with reals clients over at least one year, that's the proof, not "just" winning a hackathon.
Ingo, that matches what I see too . some sessions the agent finds the right pattern in mins, other sessions I spend hours checking its output line by line. The line you draw between winning a hackathon and running production for real clients over a year is the test I keep coming back to.
My submissions this month will look fine to judges over a weekend. Whether any of them survive a paying client past month 3 is a different question and I don't have an answer yet for most of them.
Wonderful work with your article!
I get it too
Thanks, Ali...
Fast generation changes what a hackathon can reward. It may be worth adding judging signals that are hard to fake with a polished prototype: a clear problem constraint, a live failure mode, or evidence of user feedback. That keeps the event focused on judgment and learning rather than output volume.
Hackathons could score understanding separately from completion. Require design notes tests and maintenance plans so AI speeds building without hiding decisions.
Justin, that is a solid idea. Score design notes and maintenance plans not just working demos.
Vibe-coded submissions lose that edge fast. My worry: judges get one weekend.
Can they read a maintenance plan in that time? Maybe the judging format needs to change too.
Some comments may only be visible to logged-in visitors. Sign in to view all comments.