Vibe coding a game jam entry in a weekend

Do this before anything else, because it can decide whether this tool is usable for your jam at all. Many jams require a browser-playable build at a URL so judges can try entries without installing anything. Flockbay builds a game that runs locally on the copy of Godot bundled with the app, on Windows. If the rules say web-playable, that is a hard mismatch and you should pick a different tool rather than find out on submission night.

Jams that accept a downloadable PC or Mac build are a straightforward fit. So are jams judged from a video, and internal or informal jams with no distribution rule at all.

The other rule worth reading is the one about AI. Some jams require disclosure, some cap how much of the entry can be generated, and a few ban it. Flockbay writes the GDScript in your project, so you are within scope of any of those rules and you should know which one applies.

More on this: The 2026 Cursor Vibe Jam: what the rules asked, what won, and how to prepare for the next

The jam entries that finish are the ones with one mechanic and one place. Not one mechanic and a world — one mechanic and a place you could describe in a sentence: a flooded quarry, a single street at night, one long staircase.

That shape suits how this tool divides the work. The place is the part you build by hand, and hand-building one strong location in a few hours is realistic. Building three is not.

Spend the first block on the world, not on behaviour. Once the place exists you can feel what the game should be in it, which is faster than deciding in the abstract and then trying to build a world that agrees with you.

Which world depends on an answer you give before anything is created: a new project asks 2D or 3D, as two pictures, with no default. In 3D the block goes on terrain, water, weather, the buildings and where the light falls. In 2D it goes on dragging to say where the ground is and dropping sprites onto it — and it starts further along, because a 2D project is playable the moment you make it. Sky, a band of scenery, ground with a ledge and a floating platform, and a person who walks, falls and jumps under the keys. On a jam clock that is the whole of the first evening, already spent for you.

More on this: What is vibe coding?

Give Flockbay one change, then play it. The temptation under time pressure is to fire off five requests and wait, and it reliably costs you more time than it saves, because when something is wrong you now have five candidates for why.

Be specific in the way you would be with a person. "Enemies should stop chasing once I am two rooms away, then go back to patrolling" is workable. "Make the AI better" is a request for a guess.

The game runs locally when you press play, so testing costs you nothing but the seconds. Under jam pressure that is the difference-maker: play constantly, and let the build tell you what to do next rather than the plan you wrote at hour one.

Ordinary jam advice, and it holds here. The last block goes on the beginning of the game, the readability of the goal, and the one bug that will make a judge quit. Not on a new mechanic.

Build early, not at the end. Make a build at the halfway point and run it, so that packaging is a known quantity rather than a discovery you make at 2am.

Write your AI disclosure while you still have your notes. If the jam asks what was generated, "Flockbay wrote the GDScript; the world was built by hand in the editor" is an accurate and easy sentence to give them.

Flockbay is a desktop app for Windows and it has to be installed. Do that before the jam starts, along with making a throwaway project, so that hour one is not setup.

You need an internet connection to build. The chat, asset generation and world data all come over the network. Playing what you have already built does not need one, but a jam without connectivity is not a jam you can do here.

The app is free, and a fine way to try the loop. Flockbay is free on your own AI: your Claude or ChatGPT plan, your own agent, or an API key.