I Handed My Website Over to AI Agents: An Audit Revealed 25 Useless Edits and 4 Hallucinations

The author shares their experience using AI agents for the full lifecycle of website development and maintenance. With over 200 commits made, the primary challenge shifted from writing code to verifying it. The speed at which AI generates changes significantly outpaces human capacity to review them, creating risks of deploying low-quality or non-functional code. An audit of 111 changes proposed by the agents revealed that 25 were useless, and 4 were pure hallucinations. The author emphasizes that automation without integrated, rigorous quality control leads to a scenario where code only 'looks like it works,' masking potential errors. The article highlights the critical need to adapt development processes when transitioning to autonomous systems, where human oversight of AI output becomes the new bottleneck.