PwC builds AI agent to reason over enterprise spreadsheets

PricewaterhouseCoopers has built what it calls a “frontier AI agent” designed to do something that most corporate automation pitches conspicuously avoid: survive contact with real spreadsheets.

According to Business Insider, PwC’s in‑house engineers developed an agent that can “reason over vast, enterprise-grade spreadsheets”—the kind that behave less like worksheets and more like sprawling, interlinked financial machinery. Matt Wood, PwC’s global and US commercial technology and innovation officer and a former AWS AI executive, told Business Insider that conventional AI tools “just kind of shrug and give up” when confronted with multi‑tab workbooks containing millions of cells plus charts, images, and scanned receipts.

Technically, PwC is aiming at a hard problem: reliable tool use over complex, structured artifacts. Spreadsheet work is not primarily about eloquent text generation; it is about navigation, dependency tracking, reconciliation, and defensible transformations. An agent that can scan across tabs, chase references, extract evidence, and perform consistency checks is essentially an LLM orchestrator wrapped around spreadsheet parsers and deterministic tooling—then packaged with the controls Big Four clients demand.

Those controls are the real product. “Enterprise-grade” in this context implies access control, permissions that mirror client data boundaries, audit logs that can survive discovery, and guardrails against accidental (or convenient) data leakage. If PwC can make the system produce repeatable outputs—what changed, why, and based on which cells—it turns spreadsheet labor from artisanal priestcraft into something closer to a governed pipeline.

Wood claims the payoff is dramatic in audit walkthroughs: what used to take weeks of manual evidence gathering across sprawling workbooks can drop to hours once the agent maps structure, extracts relevant data, and runs validation checks, Business Insider reports. That is plausible, and also slightly ominous for the armies of junior staff whose job has been to click through tabs until the numbers stop screaming.

The more interesting shift is organizational. PwC recently launched a dedicated engineering career track, signaling that the firm wants to be a software builder, not merely a billable-hour interpreter of other people’s software. The Big Four have long monetized bureaucracy by managing it for clients; productizing that bureaucracy into tools is the logical next step.

There are reasons to be skeptical of what happens when compliance factories industrialize. A spreadsheet agent can reduce wasteful manual work—but it can also make surveillance, documentation, and “controls” cheaper to impose. When a firm that profits from auditability builds the machine that generates the audit trail, the incentives are… beautifully aligned.

Still, if anyone is going to automate Excel hell, it might as well be the people who created so much of it.