*A field manual to build your ambitious ideas faster with AI, alongside the most talented technical individuals of your city, in a "half-hackathon" format.* Hackathons used to be cool. But now with AI, *hackathons are so back*. I grew up going to hackathons. And the new hackathons I'm going to are shorter, and a lot more ambitious. These new hackathons are in fact so different, **a new category is emerging**. They are "half-hackathons". Why? Participants can build something impressive ***in half as much time***. And this has implications: First, the bar is getting higher. In one prompt and 3 minutes, you can build a working software in Lovable. So why not try to shoot higher? Make something you can actually use by the end of the 6 hours? Try launching dozens of parallel agents and *make novel scientific discoveries*. An arrogant claim? No, this is now possible. ^1 Second, this now means hackathons are becoming the best way to explore new ideas quickly, from -1 to 0. For testing an idea, you don't need 3 month incubators or 24h hackathons anymore, a half-hackathon is all you need. They give us a time to explore in favorable conditions, with a deadline, peer pressure for demos and builder energy. What sets half-hackathons as an entirely new category? - **Shorter time frames:** half-hackathons are shorter than hackathons. Between 4 to 8 hours. From experience, 6 hours seems to be a good length and gives time to chat with others before kickoff and after demos. They can fit in an evening (after work), or an afternoon. Smaller time commitments means more potential talents can participate. Yay! - **Less code, more code:** Teams focus on their "Plan.md" written in plain text, in markdown format (PRDs, etc), delivered to agents that write the code. This is an important distinction. Time spent actually coding is reduced. Participants actually talk more than in normal hackathons. Why? Because we have more time to chat in between agent requests sent. - **Parallelization of thought**: the point above also means for teams to be able to explore multiple paths in parallels. This is a powerful mindset shift. AI power users launch multiple experiments instead of building sequentially. - **Templates and boilerplates:** It's now easy to grab a boilerplate from a public repository and get started with a fast proof of concept. And the good news: these templates have exploded in number in the last decade. They're perfect for half-hackathons. - **Smaller team sizes:** Teams of 4 or 5 people used to be necessary to get impressive results. But they're slower. Now tiny teams of 1 to 3 people can ship more. - **Better for building product and pitch skills**: AI-powered development workflows require builders to clearly state what they have in mind. They need to have a clear intention and communicate it to the machine (prompting), and to others (pitching). So the quality of demo presentations is higher (especially when LLMs can give you pre-feedback, or help you find a great pitch for your demo). - **Democratization beyond coders**: just like half-marathons attract more people than full marathons, the same seems true with half-hackathons. Shorter time-frame, with no-code AI tools, means a lot more semi-technical people can benefit from hackathons. Designers, marketers and product managers can now contribute, widening the talent aperture. ### Joining a half-hackathon Lu.ma is now the new default website to find interesting hackathons in your city. No hackathons in your location? See the section below. Once you found one, I built this [GPT guide](https://chatgpt.com/g/g-688a35683b708191a9e9e8909200db66-half-hackathon-participant-guide) to help you make the most of it. Below is a quick preview. Making the most of your 6 hours is finding a good equilibrium between diverging on different ideas, then converging on the most promising one. Diverging and converging in short loops. This is a suggestion for how to best ship in 6: ![[Team Collaboration Guide on Chalkboard.png]] ### Forming a team Half-hackathons are the easiest way to test a collaboration with some of the most talented technical people in your city. Make use of it! Tiny teams that learn from other’s rhythms can spin up a project faster than individuals. If you find great teammates, keep them across events; that continuity builds compound skill. Have your own group chat, brainstorm, meet after hackathons, have a drink. Aim for one half-hackathon per month. Every quarter, join bigger/longer hackathons. Regular gathering will help you bond faster and iterate quicker towards your ideas. ### Your turn I'm taking time in the coming months to try to restore the dignity of hackathons, and amplify their impact with their 'half' format. I can't find a better way right now to be helpful to others than promoting these types of events. Now it's your turn: join a hackathon, find a team, then organize your own half-hackathon. Here's a guide I wrote: [[Organizing a half-hackathon]]. Shoot me your city, date and theme and I'll share the event to the network. I'll send you extra help too. Reach out on [𝕏](https://twitter.com/ben_issen) or by [email](mailto:[email protected]). Let's make hackathons cool, ambitious, energizing, and crazy again. ![[Pasted image 20250730220504.png]] Thank you Julian Blum, Salim Boujaddi, Krishaan Khubchand, Timothée Issenmann for reading drafts of this post and suggesting edits. Beni July 2025 ^1 A month ago, I went to the most impressive hackathon of my life. I flew to San Francisco, and participated in a hackathon organized by Anthropic and Exa. Here was the pitch: "​Can AI agents discover new science? We believe these systems are now smart enough to legitimately make novel discoveries. Join the challenge and get to be the first in human history to demonstrate something new." ([Full version](https://lu.ma/sikose73?tk=zSgM1U)). An arrogant claim? No, it [worked](https://x.com/ExaAILabs/status/1931820487363142069). And I left the hackathon realizing we had entered a new era. ![[Half-hackathon-promises.png]]