This summer at Interacty we ran our first hackathon. It wasn’t about who could code faster. Instead, we designed a game: two hours, a hidden task for each person written on a card. Employees should have solve it with ChatGPT or any other LLM. At those time, the 4o model was the top model available and felt like the most advanced tool in the world.
What happened was unexpected and not expected at the same time. After a quick intro, the team jumped in. Some had coding backgrounds, others had never touched production code. Yet when the demo round came, the results looked almost the same. Engineers and non-engineers delivered projects at the same quality level. That was the first nottable thing.
The second surprise was less fun: the gap between hackathon magic and production reality. The demos were creative, but unstable, fragile. Hallucinations, unstable outputs, and isolated pieces of code made it clear—we couldn’t just “drop AI in” and expect it to work.
Looking back two months later, the biggest lesson is that AI adoption isn’t just a tool problem. It often demands a business refactor. You can’t simply put AI on top of existing company processes. You need to rethink the process itself.
We now have a clearer vision of where AI fits into our core systems. But it will take extra efforts to move from playful demos to reliable, production-level value. The hackathon was just the spark and first step to true AI-first company.
Artem Grishanov