Chat and citations won't save your vertical AI - Atul Ramachandran, Filed Inc

Vertical AI products often fail to deliver on the promise of saving time and money because they rely too heavily on synchronous chat and manual verification via citations. To truly automate complex workflows in sectors like healthcare, legal, and tax, developers must shift from designing for user participation to designing for agentic delegation. This model treats the AI as a worker on a conveyor belt, where the user acts as a supervisor who monitors progress, provides specific skills, and intervenes only when necessary. Success in this paradigm requires building robust features for task delegation, monitoring, and control, while prioritizing "Weekly Active Sessions" over "Weekly Active Users." By allowing agents to execute long-running tasks independently and providing users with the ability to audit and override decisions, products can achieve higher efficiency and build greater trust.