From AI experimentation to AI execution.
Teams build multi-agent AI, networks of specialized agents that coordinate to automate real-work tasks, so people can put more focus on the decisions that matter, and get more done.
This year
The 4th edition raises the challenge.
OQBi's hackathon returns for its fourth year, and this time participants team up to build multi-agent AI: networks of specialized agents that coordinate to automate real-work tasks, freeing people to focus on decisions and lift productivity.
Past editions were about discovering what AI could do. This year is about making it work, shipping agent systems that carry a real workflow end to end, not demos that stop at the first hard step.
Every build answers to two rules: at least three coordinating agents and one external database as the shared source of truth. The challenge section below shows fifteen worked examples across five sectors to get teams thinking.
Less prompting-as-magic, more systems that run. Teams are judged on solutions that actually do the work, coordinated agents, real data, an output someone would use on Monday.
Schedule
Key dates: September to October 2026.
The challenge · idea starters
Pick a workflow. Then split the work across agents.
Every example below fits the two rules — at least three agents and one external database — and each is broken down the same way, so you can see why more than one agent earns its place.
The pattern to notice: one agent reads, one reasons, one acts, and sometimes one verifies. When those jobs are genuinely different, multi-agent isn't decoration, it's the design.
These are starting points, not tracks.
The strongest submissions usually mutate one of these into something narrower — a specific role, in a specific team, doing a specific thing that's painful today.
A good test before you build: can you name a real person who has this problem, and say in one sentence why one well-prompted agent wouldn't be enough? If you can't, tighten the idea before you write code.
Registration is open
Team up. Build the agents. Take the stage.
Register to join Hackathon 4.0 — start with debate training, compete to be shortlisted, then build your multi-agent solution over the two build days.