After incidents, we organize “Lessons Learned” sessions. One of the teams put as incident root cause something close to: “It was asked by my Lead to do this”.

The agentic coding loop is the trend of the moment. I think it's the wrong default for most coding work. The issue isn't AI itself; it's what you give up when you hand over too much.

MCP standardizes how LLMs interact with tools and data, but it doesn't cover interactive UI. MCP Apps fill that gap.

I went to Techorama on May 12 and 13. Here’s what I learned from the AI sessions I attended.

Coding agents don't struggle equally across all types of software work, but why? I think the reason isn't complexity but feedback loop quality: How far an agent must reason beyond the immediate code to verify its own output.