There was a time when winning a hackathon required something vaguely resembling computer science. You needed to know how your application worked, how your backend communicated with your frontend, why your database wasn’t on fire, and occasionally, what the hell a WebSocket was.
Those were simpler times.
Today, you need three things: a decent prompt, a collection of AI subscriptions that would make your bank account file a police complaint, and enough design sense to make a generated landing page look like it belongs in a YC pitch deck.
Welcome to the modern hackathon.
The problem isn’t that AI is being used. AI-assisted development is fantastic. Having an LLM generate boilerplate, explain an obscure API, find a bug, or save you from writing another 400-line React component is objectively useful. The problem begins when the distinction between building something with AI and having AI build something for you becomes irrelevant to the judging process.
A few years ago, a hackathon was at least partially a test of engineering ability. You had a ridiculous idea, 24 or 48 hours, questionable amounts of caffeine, and a laptop that sounded like a jet engine. You had to figure out how to turn your idea into something that actually worked. You’d spend hours debugging an API, fighting with authentication, wondering why your database was returning undefined, and eventually discovering that you had forgotten a single await.
Now you can describe the same application to an AI coding agent in roughly the time it takes to boil water.
“Build me a SaaS platform that uses AI to analyse student productivity, with authentication, real-time collaboration, analytics, a chatbot, a modern responsive UI and a PostgreSQL backend.”
And congratulations. You are now a CTO.
This creates an interesting problem for hackathon judging. Two teams can submit applications that look almost identical in quality. One might have spent the weekend designing the architecture, implementing the backend, debugging race conditions and understanding every component. The other might have spent the weekend refining prompts until an AI agent produced something vaguely functional, then spent the remaining six hours adjusting gradients.
From the outside, both look impressive.
Ask the second team what a WebSocket does, however, and suddenly the WebSocket becomes a very philosophical concept.
Then there’s the economics of AI tooling. One participant has a free-tier chatbot. Another has ChatGPT, Claude, Cursor, Copilot, an AI coding agent and enough model credits to make a small cloud provider nervous.
At that point, the hackathon isn’t simply measuring who can build the best solution. It’s partially measuring who has access to the best development infrastructure.
And yes, “just use the free tools” is technically an option. So is running Linux on a Pentium 4. That doesn’t mean the experience is identical.
The strangest part is what happens when judging prioritises presentation over engineering. A polished application with a beautiful dashboard, animated cards, an AI chatbot and a presentation containing the words “revolutionising the future of X” can be incredibly convincing.
Whether the application is actually well engineered becomes almost secondary.
You can have a team that understands distributed systems, wrote their own backend, designed a sensible database schema and can explain every architectural decision. You can also have a team whose entire technical stack is essentially “we asked Claude.”
If the second team has the better demo, better slides and a more convincing pitch, guess who gets the trophy?
We’ve reached the slightly absurd situation where you can watch a team present an AI-generated application, ask a basic technical question about how their own system works, receive an answer that sounds suspiciously generated five seconds ago, and then watch that project win anyway.
At some point, you have to wonder what exactly the hackathon was testing.
Presentation skills? Prompt engineering? Subscription management? The ability to make an LLM hallucinate an entire software architecture and then pray the demo doesn’t crash?
The irony is that AI hasn’t made engineering irrelevant. It has made engineering knowledge more valuable, while simultaneously making it harder for hackathons to distinguish between someone who possesses that knowledge and someone who simply knows how to ask a model to imitate it.
Hackathons used to reward the ability to build something under ridiculous constraints. Now, increasingly, they reward the ability to produce the appearance of having built something under ridiculous constraints.
There is nothing inherently wrong with using AI. Hell, I’m using AI too. The irony of writing an article complaining about AI while AI is sitting somewhere in the writing pipeline is not lost on me.
But there should still be a difference between using a tool and outsourcing the skill the competition is supposed to evaluate.
If the winning team cannot explain their own architecture, debug their own application, or answer basic questions about the technologies they supposedly built with, perhaps the trophy belongs to the AI model.
The humans just happened to be standing nearby.
The moral of the story is thus: Hackathons used to ask, “What can you build in 48 hours?”
Now they increasingly ask, “How many AI subscriptions can you fit into 48 hours?”