A QCon talk on why coding agents fail: bloated context, not weak models

A QCon London talk argues coding-agent failures usually trace to bloated, noisy context rather than model quality, and lays out fixes.