In her QCon London keynote, Birgitta Böckeler, Distinguished Engineer at Thoughtworks, discussed the evolution of AI-assisted coding over the past year, highlighting a shift from “vibe coding” to autonomous coding agents. She identified increasing security concerns and rising costs in agent-based development as significant challenges. Böckeler emphasized the importance of context engineering, which has evolved from a monolithic rules file to a more granular approach with “lazy loading” capabilities to refine coding results effectively.
As AI agents can now operate unsupervised for extended periods and integrate with CI/CD pipelines, the potential for errors has grown. Weekly security incidents tied to prompt injection demonstrate the risks involved. Böckeler proposed a risk assessment framework based on mistake probability, error impact, and detectability. She concluded that while technical advancements are notable, the real challenge lies in establishing effective practices for coding agents, cautioning that AI will magnify existing developer practices, both good and bad.
Source link
