Agents Gone Rogue — AI Hackathon

SAAF (Shared Audit Agents Framework) brings together professionals from 65+ organizations to build open-source, vendor-agnostic AI agents for audit, risk, and governance. Twelve sessions into 2026, the approach has stayed simple: think big, start small, move fast. Shared pain points became reusable patterns, patterns became working agents. Now the focus shifts to making those agents reliable, explainable, reusable, and safe. That's what this hackathon is for. Developers, AI builders, auditors, and creative problem-solvers spend the day pushing SAAF's agents further. What comes out of it feeds directly into the Shared Audit Agents Framework, set to publish at the end of the year, backed by NOREA, IIA, ISACA, NBA, and ForHumanity. Co-creation, not competition. Build, share, improve. Hosted by SAAF Project and Just Eat Takeaway.com, with TAG as partner. Speakers Mathijs Schouten and Eduward van de Kamp (Mollie), founders of SAAF, open the day. Mathijs talks about AI conquering the world, self-replicating code, and mapping human behavioral traits onto AI models. Annarie (Crause) Oosthuizen, Global Head of Risk & Audit at Prosus and Naspers Group, presents ToqanClaw — building any business tool from a single conversation. Job Angula (Just Eat Takeaway.com) and Mathijs Schouten introduce the use cases teams build on today. In the afternoon, James Osborn covers AI at Just Eat Takeaway.com. Schedule 09:00 — Doors open. Registration, coffee. 09:15–10:30 — Opening talks: Eduward van de Kamp and Mathijs Schouten (SAAF founders) on agents gone rogue, Annarie (Crause) Oosthuizen on ToqanClaw, and Job Angula & Mathijs Schouten on today's use cases. 10:45 — Build time begins. 12:30 — Lunch, included. 15:00 — James Osborn: AI at Just Eat Takeaway.com. 17:00 — Submissions close. 17:00–17:30 — Judging. 17:30–18:00 — Winners announced. Drinks follow. The brief Improve an existing agent. Build a reusable skill. Create better guardrails. Automate a quality check. Or bring something we haven't thought of yet. SAAF's agents need to get more reliable, explainable, reusable, and safe. That means skills that can inspect and enrich agent repositories: better documentation, README files, instructions, controls, testing, and permission boundaries. Pick a problem in an existing agent or repo and build something others can reuse. Bring your laptop. Pick a problem. Build something that others can reuse. Hard rule: build time ends 17:00. No exceptions. Submit all three: a working demo judges can try live, a 60–90 second video showing what you built and how it works, and a public GitHub repo. Judging criteria We're looking at creativity, inventiveness, contribution to agent quality, and, most importantly, execution power. Concept & Originality — 40%. How far did this go toward improving quality control? Technical Execution — 30%. Do the tasks the agent performs complete, and are the activities logged for re-performance? Real-World Relevance — 30%. Does the agent perform real, working tasks, or a script pretending to? Judges Judge 1 — Mark Poen Judge 2 — Mathijs Schouten Judge 3 — Eduward van de Kamp Prize A small thank-you for the strongest contribution: €500 in Just Eat Takeaway.com vouchers. Who's behind it SAAF Project is a free community where auditors from 65+ organizations build AI audit agents together, in the open. JET (Just Eat Takeaway.com) is a global food delivery company turning into a tech company, using AI across logistics, restaurant onboarding, and voice ordering. TAG is an Amsterdam community of builders and hackers. Founded 2024. Ships every month.