Grona’s hackathon let synthetic customers pick winners

Synthetic customers ran the room. On the Sunday afternoon just before LEAP kicked off in Riyadh, a group of marketers gathered at COLABS Narjis to build end-to-end campaigns using agentic AI. But instead of pitching their work to a panel of judges, they handed the verdict over to an automated funnel. Around seventy people showed up for the AI Marketing Hackathon, backed by Google for Startups credits and put together by Grona, COLABS, and Salesflo. The assignment was simple: ship a working campaign for an actual product, not another slide deck about agents. By 15:30, laptops closed, and roughly a thousand synthetic buyers, built to mirror each team's stated target audience, walked the entire simulated journey from start to finish. That was the real punchline. Everyone talks about AI agents, but almost nobody has actually put one to work on real tasks. The line went up on the projector early in the day, and it sums up the state of marketing right now. We have had two years of shiny product demos, Twitter threads, and LinkedIn breakdowns. They made people fluent in the buzzwords, but they rarely led to someone running an agent on a live brief. Hype thrives when an agent can spit out ad copy, images, and landing pages without ever having to convert a real stranger. This room tried to fix that by replacing the human judges with simulated buyers. The scoring was straightforward, public, and brutally honest. Half of it came down to conversion rates across that simulated journey. Another thirty percent tested how well each team's agent held up against a brief they had never seen before. The final twenty percent measured whether the campaign felt like a real, cohesive user journey or just a rushed prototype. Charm could not help you here, and stage nerves could not hurt you. A thousand synthetic customers do not applaud. The turning point came right after Zuhr prayer. Most of the marketers in the room had never touched an IDE before. A quick masterclass walked them into VS Code and through a repository of sixteen pre-built agents, focusing entirely on the one skill that actually counts: writing instructions specific enough to get reliable results. Together, the room built a base marketing agent before splitting off into teams. Nothing in the stack was hidden. The tools were public, and the code was open to download. All people really needed was the permission to try, and an hour with someone nearby to help when things got stuck. There was no big trophy ceremony at the end, which fit the spirit of the event. The point was never a staged photo op. It was finding out whether a marketer who spends their days writing briefs for other teams could sit down and, in five hours, build something an artificial buyer would actually complete. Salesflo brought the practical operator mindset, COLABS provided the space, and Grona gave the room a reason to test the limits. The real story here is the judging format, not just another side event during a busy conference week. Riyadh calendars are packed with keynotes and high-level panels, but this afternoon asked Gulf marketers, agency leads, and founders to put their work on the line against an audience built from their own data. That looks a lot closer to what running a modern Saudi e-commerce brand actually feels like. If this practical format makes its way to Jeddah and Dubai without turning into a standard sponsored vendor showcase, it could change how teams build here. If it stays a one-off afternoon in Narjis, it was still a great workshop. A few realistic caveats belong in the conversation. A synthetic audience is only as accurate as the profile the team feeds into it. Simulated clicks are not the same as live ad spend, and a five-hour sprint cannot prove long-term revenue impact. Free cloud credits make the first test cheaper, but they do not replace a real customer pipeline. The real payoff starts after the event. If a few of those teams took what they built and plugged it into a live campaign the following week, the hackathon did its job. If the code stays tucked away in a folder and never touches real traffic, the gap remains. The real test was never the score on Sunday. It is what happens when real budgets hit a live funnel.