Can AI Agents Run Entire Business Processes Alone?

Can AI agents solo entire processes like order-to-cash? Partially yes, fully? Not quite—but closing fast. End-to-end pilots at Brex handle invoicing: agent ingests orders, checks inventory via ERP API, generates invoices, chases payments, updates ledgers. 90% autonomous, humans for exceptions.

Strengths: Parallel processing—agents swarm subtasks. Tools like Retrieval-Augmented Generation pull accurate data. Weaknesses: Black swan events (e.g., supply shocks) demand human intuition.

Procure-to-pay at P&G? Agents scout vendors, negotiate via email bots, approve spends. Success rate: 85%. Scaling to full biz processes needs orchestration layers (e.g., LangGraph) for error recovery.

By 2028, expect 70% coverage in routine flows. Barriers: liability (who’s liable for agent errors?) and data silos. But with chain-of-thought and self-critique, they’re there. Revolutionize or risk obsolescence.