Madeline Minshew describes the gap between a promising agent demonstration and a workflow that teams can trust. Shared context, security and coordination become central when AI systems cross organizational boundaries. Better model capabilities do not automatically resolve these operating problems.
Tony Conrad contrasts executive-led AI adoption with earlier bottom-up software purchasing. He describes companies embracing AI quickly and only later asking whether their model bills produce enough value. His account of one founder’s unusually large Anthropic bill is an anecdote, not audited evidence of typical spending.
Tony Conrad suggests using capable frontier models for difficult initial work and considering cheaper models for repeatable tasks where they perform adequately. Both speakers emphasize that adoption requires changes in behaviour and workflow; software seats and security reviews do not disappear simply because agents improve.
Watch on YouTube
