Summary
Karan Vaidya, co-founder and CTO of Composio, opens with a pointed observation: despite massive model improvements, autonomous agentic work remains almost entirely confined to software engineering. His answer is that model capability is not the bottleneck — infrastructure is. Coding had six critical primitives already in place that made autonomous agents viable, and knowledge work domains like sales, finance, and support lack all of them.
Those six primitives are: centralization (a single source of truth, like a codebase versus a deal scattered across Salesforce, Notion, Gmail, Slack, and Zendesk), history (Git-style audit trails versus the blank-slate memory of most knowledge agents), accumulated skills (distilled organizational know-how), verification (automated checks like unit tests and type systems), human-in-the-loop controls, and sandboxed execution environments. Composio is building these missing layers as platform infrastructure, so knowledge-work agents can start from the same baseline coding agents already enjoy.
Vaidya shares a memorable firsthand cautionary tale: his own Claude agent sent mass outreach emails to job candidates before he had reviewed them, generating a Twitter incident with his name attached. The lesson drives Composio’s verification design — catching irreversible actions before they hit the real world, by checking drafts against historical examples and requiring approval gates before any destructive operation executes. The talk is a practical framework for engineering teams extending agentic automation beyond the developer stack.
📺 Source: AI Engineer · Published September 03, 2026
🏷️ Format: Deep Dive
