3rd Copenhagen Symposium on Human-Centered SE AI
Adding to Daniel Russo’s reflection on the third Copenhagen Symposium on Human-Centered Software Engineering AI: In my view, this third iteration was the most focused and action-oriented yet.
The first symposium was largely exploratory, figuring out the role of SE in a world of GenAI, and asking ourselves what our community can still contribute. The Copenhagen Manifesto came out of that, and its principles are therefore more fundamental than actionable.
The second iteration introduced workstreams. One of them produced the LLM guidelines project, which is finally approaching a stable state (with continued maintenance ahead, of course).
In this third iteration, the group spent time debating whether the Copenhagen Manifesto needs refactoring, or even a full reengineering. At the same time, several new workstreams were formed.
My main takeaway of the discussions: SE is as relevant as ever. A lot of the innovation right now is happening around the agent harnesses rather than the models. One example of this: On the train this morning, I read Daniel Stenberg’s post on Mythos and curl. Mythos surfaced just one low-severity issue, because earlier AI scans had already caught high-severity issues. Building a harness that spends millions of tokens scanning every file of a project was already feasible with the previous generation of frontier models, and it was (and is) effective.
The frontier model is one component of a larger software system. Where it sits in the pipeline, what it sees, how context is curated, and how it interacts with tools determine what one can actually achieve with the model’s capabilities.