PwC and OpenAI announced a partnership to build AI agents on corporate finance workflows: planning, forecasting, reporting, procurement, payments, treasury, tax, accounting close. OpenAI plays "customer zero", meaning it uses the agents internally to validate governance and runtime controls before bringing them to PwC clients. One number comes out of the announcement: Codex processes 5x more contracts with the same team.
Anyone who has run an admin function knows what this means. Monthly close isn't an intelligence problem, it's a volume problem on repetitive tasks: bank reconciliations, PO-delivery note-invoice matching, VAT control on intra-EU purchases. Taking that volume off the controller's agenda frees up hours spent today chasing variances of a few euros. The real point of the PwC-OpenAI partnership isn't the agent itself, it's that the Big Four wraps its own control framework around the agent. For a CFO who has to answer to the board of statutory auditors (collegio sindacale), it's the condition for the piece not to stay a POC.
Why this matters for anyone building enterprise AI: the agent's governance is the product, not the agent.