HackEurope 2026 ran as a 30-hour overnight hackathon across Dublin, Paris, and Stockholm on February 21–22, and its judging rubric on Devpost asked a blunt question about every submission: is this "a complete, well thought out package, or stitched up work by 10 AI agents?" That single line is the real story of AI and hackathons right now — organizers want AI-assisted builds, just not AI-assembled ones.
Short answer: HackEurope 2026 (Dublin, Paris, Stockholm; Feb 21–22, 2026; 760 registered, €50,500+ in prizes) didn't ban AI tools, but its judging criteria explicitly penalized projects that read like "10 AI agents" stitched them together overnight. The lesson for any hackathon: use AI to move faster on boilerplate, not to replace the thinking judges are actually scoring.
I test AI coding assistants — Copilot, Cursor, Claude Code, ChatGPT — against tight deadlines for a living, and the HackEurope judging line matches exactly what I've seen happen when a team leans on AI too hard under a clock. In my testing, an AI assistant can get a team from a blank repo to a working prototype in an hour that used to take an afternoon. It cannot get a team a win if nobody on it can explain why the code works. Here's the actual rant, plus a practical way to use AI at a hackathon without tripping the "stitched up by 10 agents" alarm.
What you'll need
Nothing exotic: a laptop, a GitHub account, and one AI coding assistant you already know how to drive — not one you're learning for the first time at 2 a.m. HackEurope's own terms and conditions allow "pre-existing libraries, frameworks, and APIs" and any sponsor-provided tools, and organizers told participants unsure about AI-assisted development to email the team directly rather than guess. That's the norm at most 2026 hackathons: AI tools aren't banned, but the rules rarely spell out exactly where the line sits, so you're expected to use judgment. Bring a free-tier account for ChatGPT or Claude, and either GitHub Copilot or Cursor set up before the clock starts — not mid-event, when setup time eats into build time.
Step-by-step: using AI at a hackathon without it working against you
1. Read the judging rubric before you open an editor
HackEurope scored on three things: potential to become a startup, technical complexity, and execution quality. None of those reward volume of AI-generated code — they reward a working, coherent product. Know what you're actually being scored on before you decide how much of the build to hand to an assistant.
2. Pick one AI tool and commit to it
Switching between ChatGPT, Copilot, and Cursor mid-hackathon costs more time than any of them saves. Decide going in: Cursor or Copilot for in-editor completion, ChatGPT or Claude for architecture questions and debugging you paste in manually.
3. Use AI for the boring 60%, not the differentiating 40%
Boilerplate, config, CRUD endpoints, a starter README — that's where an assistant earns its keep. The feature that's actually your pitch — the part a judge will ask about — you should write and understand yourself, even if AI drafts a first pass.
4. Keep a build log as you go
A running note of what you built, what you asked AI to generate, and what you changed by hand does two things: it makes your 2-minute demo video easier to script, and it's your answer if a judge asks "did you write this?"
5. Test the demo path by hand, not by prompt
AI-generated code compiles more often than it works end-to-end. Before you present, walk your actual demo flow — click by click — without asking an assistant to "check if this works." Judges will run it live in their heads; you should have run it live for real.
6. Be ready to explain any line judges point at
This is the practical version of HackEurope's rubric line. If a judge asks about a function and nobody on the team can explain it, that's the "stitched up by 10 AI agents" outcome, whether AI actually wrote 10% of the code or 90%.
Example prompts you can copy
These are the prompts I reach for when testing AI coding tools under a deadline — they work for hackathon builds because they ask for a starting point, not a finished decision:
- "Scaffold a Flask API with a
/predictendpoint that accepts JSON and returns a mock response — I'll fill in the model logic." - "Here's my Express route [paste code] — it throws a 500 on empty input. What's the likely cause, and what's the minimal fix?"
- "Write a 3-sentence pitch summary for a hackathon demo: [describe the product], aimed at judges scoring on startup potential and technical complexity."
- "Generate a README with setup steps, a one-line project description, and a short architecture summary for this repo: [paste file structure]."
- "Review this function for edge cases I might be missing before a live demo: [paste function]."
Each one asks AI to do a bounded task you can check in seconds — not "build my hackathon project," which is exactly the prompt that produces a stitched-together mess.
Common mistakes to avoid
The biggest one I see, in testing and from teams I've talked to: nobody on the team can walk through the codebase because everyone treated AI as the sole author instead of a fast first draft. Second, teams generate a slick-looking UI with AI and skip testing the actual logic underneath — judges notice the moment they click something the demo video didn't show. Third, ignoring the rules entirely instead of emailing organizers when AI usage is ambiguous, which is exactly what HackEurope told participants to do. Fourth, over-indexing on one AI tool for everything, including tasks it's weak at — asking a code assistant to write your pitch copy usually reads worse than writing it yourself in five minutes. Fifth, treating a 30-hour AI-assisted build as equivalent to a 30-hour build from scratch — it isn't, and pretending it is in your pitch is the fastest way to lose judge trust.
Tools that make this easier
If you're building at a HackEurope-style event, the assistant you pick matters more than people admit going in. My AI coding assistant guide breaks down what each major tool is actually good at under time pressure. For in-editor completion specifically, Cursor vs. Copilot and how to use Cursor AI cover the two most common picks at student hackathons right now. If your team leans on a chat assistant for architecture and debugging instead, GitHub Copilot vs. ChatGPT and how to use ChatGPT are worth reading before the event, not during it. And if you want the honest version of "how much should AI actually write," AI coding without the vibes is the closest thing I've written to this rant, minus the hackathon framing. My full comparison, based on testing each tool against the kind of scaffold-then-iterate work a hackathon demands:
| Tool | Price | Best for at a hackathon | Where it struggles |
|---|---|---|---|
| GitHub Copilot | $10/mo (free for verified students) | In-editor autocomplete, boilerplate, CRUD routes | Multi-file architecture decisions |
| Cursor | Free tier; Pro $20/mo | Fast multi-file edits, refactoring under a clock | Can overwrite working code if you accept diffs too fast |
| Claude (Claude.ai / Claude Code) | Free tier; Pro $20/mo | Debugging, explaining unfamiliar code, architecture Q&A | Slower than in-editor tools for small completions |
| ChatGPT | Free tier; Plus $20/mo | Pitch copy, README drafts, quick logic checks | Confidently wrong on library-specific APIs without docs pasted in |
Prices above are current standard rates as of August 2026 — check each vendor's pricing page before you commit, since hackathon-season promos and student discounts change often.
My take
The HackEurope rubric line isn't really anti-AI — it's anti-laziness, and it happens to catch a lot of AI-assisted work because AI makes laziness easier to produce at scale. In my testing, the teams that do well with AI tools under a deadline are the ones using them the way a senior engineer uses a junior one: delegate the boring parts, review everything, and never hand over the part you'd actually get asked about. Do that, and "stitched up by 10 AI agents" was never going to describe your project in the first place.
Frequently Asked Questions
Was AI banned at HackEurope 2026?
No. HackEurope's public rules and terms don't prohibit AI coding tools; the judging criteria simply penalize projects that look assembled by AI rather than built with it. Participants unsure where the line sits were told to email organizers directly.
How many people took part in HackEurope 2026?
760 people registered across the three simultaneous locations — Dublin, Paris, and Stockholm — for the February 21–22, 2026 event, competing for €50,500+ in total prizes, including a €10,000 grand prize. See the official HackEurope site and Trinity College Dublin’s recap for the full breakdown.
What's the easiest way to use AI at a hackathon without it hurting my project?
Use it for boilerplate, config, and debugging — the parts a judge won't ask about — and write or personally review anything tied to your core feature or pitch, so every team member can explain any line of code on request.
Do hackathon judges actually check if AI wrote your code?
Not by running a detector, based on HackEurope's public rubric — they judge by asking questions live and scoring execution quality and technical complexity. A team that can't explain its own project fails that test regardless of what tool wrote which line.
Which AI coding tool is best for a hackathon?
There's no single best pick — Copilot and Cursor are faster for in-editor completion, while Claude and ChatGPT are better for debugging and architecture questions you paste in manually. See the comparison table above for the current pricing and where each one struggles.