Thanksgiving Multiplayer AI Hackathon
🎟️ REGISTER FIRST (both are required)
1️⃣ AWS registration, needed to enter the building http://events.builder.aws.com/d/fczr5q
2️⃣ Join the Discord Join Discord Server
Space is limited.
⚠️ ENTRY REQUIREMENTS
18+ only
Valid government-issued physical photo ID at check-in. Digital IDs will not be accepted.
No scooters or bikes inside any Amazon building. No on-site parking.
AI has learned to talk. Now it is learning to act: to make its own decisions, collaborate with other agents, and share spaces with humans.
Multiplayer AI is a hackathon for building AI agents that behave like living characters. They move, interact, compete, cooperate, decide outcomes, and even teach humans. There is no single chatbot and no fixed script, only agents with their own goals, personalities and minds, in worlds you create.
Whether you are a developer, designer, game builder, ML engineer, product manager or storyteller, this is your chance to build what comes after chat: AI that plays, performs and teaches.
The Challenge
Most AI today is one human talking to one assistant. The next step is multiplayer: many agents interacting with each other and with humans, where the outcome is not scripted but emerges from what the agents decide.
Your challenge is to build an experience where multiple AI agents come alive as visible characters that think, act, move and make decisions, shaping a film, a game, or a real person's growth.
What to Build
Choose one of the three tracks and build a working, visual, multi-agent experience.
Every project must include:
Characters: each agent has a visual identity, such as an avatar, sprite, or 2D or 3D model
Visual output: an animated video, a running simulation, or a playable game
Movement and action: characters move, react and act on screen, not just output text
At least three agents with their own goals, roles or personalities
Visible decisions: a log or ledger showing what the agents decided and why
Projects score higher with:
3D: 2D projects are welcome, but 3D is encouraged and scores higher
ML and trained agents: agents that learn, adapt or are trained, through reinforcement learning, fine-tuning, learned behaviour, or memory that changes their decisions over time
Voice: agents that speak with their own voices
Tracks
Track 1: Agent Animated Film
Create a short animated film where the plot is written by agents, not by you. You set up the world, the characters and the opening event. From there the agents interact, make choices, and decide how the story ends.
What this track involves:
You set the world, the characters and the opening event
Agents interact and decide every plot beat, including the ending
Characters are animated, moving and reacting on screen
The same setup produces different endings across runs
The output is a 1–3 minute animated short
Examples: A heist crew with a hidden traitor, where the agents decide whether the heist succeeds. Survivors on a sinking spaceship fighting over a single escape pod. A courtroom where lawyer, witness and jury agents reach their own verdict.
Track 2: Co-op Multiplayer Game
Build a multiplayer game where humans and AI agents share the same world as equal players. The agents are not NPCs waiting for instructions. They have their own goals, memories and plans, and they act whether or not a human is involved.
What this track involves:
At least 2 human players and 3 agent players in one visual game world
Agents act independently with their own goals and memory
Agents can ally with players, trade with them, compete or betray them
At least one major world event is caused by agents alone
Examples: A survival island where agent tribes choose to welcome or betray human players. A trading town where agent merchants strike deals and hold grudges. A social deduction game where the impostor could be a human or an agent.
Track 3: Grow with Agents
The roles are reversed. Usually humans train AI; here, agents train humans. Agents step into expert roles such as interviewer, examiner, professor, coach, boss or client, and the human learns, practises, gets tested and improves through a visual, interactive simulation.
What this track involves:
Agents play expert roles such as interviewer, examiner, coach or boss, as visible characters
At least 2–3 agents coordinate about the human
Agents adapt to how the human is performing
The session ends with a clear result and evidence-based feedback
Examples: A 3D interview panel that questions you, deliberates and gives a verdict. An office simulator with an agent boss, teammates and a difficult client. A driving test with an agent examiner in a simulated city. A sales call where you pitch to a buying committee that pushes back on every claim.
Who Should Join
Developers, game builders, ML engineers, 3D artists, designers, product managers and storytellers. Join solo or in a team of up to four.
Build agents that do not just talk. Build agents that play, perform and teach.
Sponsors
AdaL
AdaL is an automation-first agent harness designed to get more done with less of your time. She combines personality, skills, and memory to become a personalized agent that can work alongside you. Check out https://adalagent.ai
Tenki
Tenki provides sandbox environments, cloud compute, and runners for executing code and workloads. It gives builders a place to experiment, run, and scale their work. Check out https://tenki.cloud
Flatkey
Flatkey is a unified access layer for AI, connecting teams to 328+ official AI models through one API. It supports models from providers such as GPT, Claude, Gemini, DeepSeek, Qwen, and more, with transparent pricing and automatic failover. Check out https://flatkey.ai
Finch
Finch provides agent skills that give AI agents reusable capabilities for performing specific tasks. Skills can help make agents more useful, structured, and consistent. Check out https://www.finchtech.ai