From coding to Knowledge work agents — Karan Vaidya, Composio

Software engineering agents have achieved full autonomy because they operate within mature, purpose-built infrastructure like version control and CI/CD pipelines. Conversely, knowledge work agents struggle because they lack these foundational systems, leading to fragmented data, lack of memory, and dangerous, irreversible actions. To bridge this gap, agents require six core primitives: centralization of data, historical logging, contextual awareness, automated verification, robust governance, and reversibility. By implementing these structures—such as sandboxed environments for testing actions and deterministic access controls—organizations can move beyond simple prompting to reliable, autonomous workflows. The current bottleneck for AI agents is no longer model capability, but the absence of the surrounding infrastructure necessary to safely execute tasks across diverse business platforms.