A VB Pulse survey published in July 2026 among 101 US enterprises found that 68% attributed at least one confident-but-wrong AI agent answer to missing or inconsistent business context over the prior six months, up from 57% in June. Recurring failures climbed from 31% to 37% over the same period. Companies with a governed context layer reported a 50% recurring failure rate, versus 21% for those without one, illustrating the detection paradox where these infrastructures make errors more visible without reducing them. Only 32% of respondents have fully deployed such a layer in production, though over 80% are engaging with this infrastructure category. RAG leads as the primary context approach at 31%, followed by governed semantic layers at 19%, and response correctness remains the top success metric at 38%.
Source: Read the original article
