Mathias Persson

I've been pair-programming with my AI agent through the CodeScene MCP for a few weeks, and it has quietly become one of those things you miss the moment it's gone. It surfaces code-quality metrics like complexity, hotspots, drift while the agent is writing or reviewing code, not days later in a separate PR review. The biggest unlock has been giving the AI a concrete view of quality. Instead of guessing whether a function is "good enough," it has structured evidence it can actually reason about and act on.

For me that's meant fewer "we'll clean this up later" moments and a steady upward pull on quality.