Caddi Debuts AI Agent for Automating Complex Back-Office Workflows

Most back-office tasks remain manual because they are defined by decision trees rather than linear steps. Alejandro Castellano, CEO and co-founder of Caddi, argues that past automation tools failed by underestimating these "fifty-branch" processes that typically reside only in the minds of veteran employees. Caddi addresses this by analyzing a firm’s existing software to identify high-cost, repetitive workflows based on actual activity data.

In the company’s "Loop Studio" interface, workers demonstrate their tasks on screen. The agent learns by observing these demonstrations, pausing to ask clarifying questions whenever it encounters ambiguity. This interactive training extracts complex exception-handling rules in minutes, a process that usually takes months of troubleshooting in traditional software deployments. Once built, the agents operate using a mix of deterministic code for routine tasks and AI for judgment-based decisions, ensuring consistency and auditability.

Early adopters are already seeing measurable gains. Palace Law has used the technology to handle a tripling of daily mail volume without increasing headcount, while The Planning Center has integrated the agent across seven workflows. The platform, which is SOC 2 Type II compliant, currently supports integrations with over 100 business applications to bridge the gaps between disconnected internal systems.