Trust and privacy identified as top barriers to agent adoption

David Pollan, creator of Assistant Benchmark, has tested over 147 AI assistants and analyzed approximately 3,600 messages from AI power user group chats. His analysis identifies trust as the number one fear, accounting for 22% of all messages discussing pain points. Pollan cited a specific incident involving Shane Mack, whose personal agent inadvertently posted financial details to a company Slack channel, highlighting the need for strict permission boundaries between personal and work contexts.

While users express concern about privacy, revealed preferences suggest a high willingness to share data for utility. For example, Hark offered $30 in pizza coupons in exchange for full access to users' Google Workspace accounts. Pollan notes that consumers have long connected bank accounts via Plaid since the Venmo era. The discussion also touches on the 'embarrassment factor,' where agents sharing social information or sending poorly toned emails creates social friction distinct from data privacy concerns. Pollan argues that solving the trust layer through better permissions and monitoring is the critical prerequisite for unlocking high-stakes use cases like finance, while noting that users prioritize powerful models over strict privacy when the utility gain is clear.