Bradesco found its first agentic-payments system slow, costly and unable to scale safely in employee beta testing, prompting a five-month architecture redesign. The bank now runs customer-facing AI across several core banking experiences for 22 million customers.
The column argues that companies should expect first-wave agentic AI deployments to need a reset and treat AI as a board-level strategic bet, not simply count pilots. It recommends staging investments as testing shows economic value, avoiding project-level demands for in-year returns and reinvesting labor and funds freed by agents into future capabilities.
The article cites IBM CEO Arvind Krishna’s account of two early missteps: building monolithic applications instead of modular components, and applying that approach in healthcare, where IBM lacked customer relationships, regulatory knowledge and clinical expertise. Krishna said acknowledging a mistake in six months is better than spending three years reaching a mediocre result.
