RoboCoders Judgement Day: From Side Projects to Software Factories

Baruch and Viktor could vibecode an app live on stage. In fact, that was their shtick for many years. But in 2027, watching two people prompt a coding agent is about as impressive as watching them use autocomplete. Stopping when the app appears to work would also demonstrate the development practice they came to argue against.

Instead, they’re bringing stories from projects that survived their first demo and had to face the proverbial “Day 2”. They’ll open the projects, inspect the agent instructions and workflows, and question each other’s fixes.

Everybody says “demos are easy, production is hard”, but what does that really mean? Viktor and Baruch will discuss familiar problems, like how an architectural decision disappears with the agent’s conversation or how the next project needs the same coding rules. Then there’s the swarm of agents, which apparently needs coordinating much like the team of software engineers it was supposed to replace. Can all of this be fixed? And what if each improvement moves the bottleneck somewhere else? Goldratt would recognize this factory.

And then the factory meets the enterprise, and things really get out of hand. Shipping something now means finding who owns the service and what else depends on it, and all that information is scattered across ten different services, hostile to AI with prejudice. Can this hot spaghetti mess become usable by coding agents, starting with ownership and dependencies they can query? And perhaps the human in the loop, put there because compliance required one, could get enough context to make a decision without serving as a rubber-stamping meat proxy. The speakers will open the workflows they’ve built to try.

Did AI solve coding? Maybe. Did AI solve software engineering? Not even close. Can we do better software engineering with AI? By the end of this session, you will be the judge.