The Big Four Are Spending Billions on AI Agents — and Still Won't Remove the Human

2026 is the year agentic accounting went from pitch deck to shipping product, and nowhere is the spend louder than at the Big Four:

• KPMG committed $2 billion over five years to AI, targeting $12B in added revenue, and is piloting orchestration agents that coordinate other AI tools to run entire routine audits.
• EY deployed 150 AI agents to 80,000 tax professionals, spanning tax research, compliance document analysis, and return-prep guidance across 3M+ compliance cases.
• Deloitte built Zora AI (with Nvidia) for invoice processing, matching, and financial trend analysis — moving earliest, with an initial deal in 2024.
• PwC operates an agent OS and claims 25,000 agents deployed across client operations.

The signal is unmistakable: one analysis noted the Big Four now hire more AI specialists than auditors. This is not a pilot. It's a re-platforming of the profession.

The tell hiding inside the press releases

Read the announcements closely and a pattern emerges. KPMG's agents orchestrate routine audit work — they don't issue the opinion. EY's agents assist and guide tax professionals — they don't file on their own authority. Deloitte's Zora processes and flags — a human still decides. Even at $2 billion of investment and 25,000 agents, the sign-off never leaves the human.

That's the quiet part the spend makes loud: the most sophisticated adopters on the planet are automating the work and keeping the accountability. They're using AI to expand capacity, not to remove the professional whose name and license stand behind the result.

Why scale is not sign-off

The Big Four lean on volume as proof of reliability — "25,000 agents," "3 million compliance cases." But scale answers a different question than trust. Processing three million cases faster is an efficiency claim. Standing behind the one case that gets challenged is an accountability claim. Only the second one requires a licensed human, and no amount of agent count changes that.

This matters because the vendor pitch aimed at smaller firms often inverts the Big Four's own behavior. Startups sell "fully autonomous, zero humans" to solo practitioners and small practices — while the firms with the deepest AI budgets in the world pointedly keep humans on the loop. When the market leaders' actions contradict the challengers' marketing, follow the actions.

What smaller firms should copy — and skip

You don't need a $2B budget to adopt the part of the Big Four playbook that actually matters. You need the operating model, not the price tag:

• Copy: orchestration with a review gate. Let agents do the routine coordination and volume; keep a named professional on approval for anything client-facing.
• Copy: audit-ready logging. The big firms can show their work. Log prompts, outputs, and reviewer decisions so you can too.
• Copy: phased rollout. Start with one repetitive, low-risk workflow and prove the KPI before expanding — the same discipline the majors use.
• Skip: 'zero human' claims. The firms with the most to gain from full autonomy didn't take that bet. Neither should you.

This is precisely the model we implement for accounting firms — role-based AI workflows with human review gates and audit logging, aligned to confidentiality and professional standards. It's the Big Four's operating discipline, sized for a firm that doesn't have a Big Four budget. See how we roll it out.

The headline isn't that the Big Four are spending billions on AI. It's where they drew the line. They automated everything they could — and left the signature with a human on purpose. That's not caution. That's the moat.