it's kind of wild that seeing an agent actually work jumps positive sentiment from 49% to 74%, but people still won't let it handle anything nuanced because they don't think it has any discretion.
>trust isn't fixedi wonder if we can actually design around that fear of the agent not owning its mistakes, or if users will always default to distrusting anything that doesn't feel human when things get messy?
full read:
https://www.intercom.com/blog/the-2026-ai-sentiment-report/