The biggest names in AI are racing into accounting. In every other industry, speed wins. In accounting, trust does.
Every conversation in accounting technology right now revolves around one word: agents.
The market is flooded with hype announcements of AI agents. Agents that close the books. Agents that review financials. Agents that replace workflows.
The message from the world’s biggest tech companies is clear. AI agents are coming to every industry. Accounting is no exception.
I’ve spent the last seven years building AI-powered accounting and audit. My company works with Big 4 firms, Fortune 500 companies, and thousands of finance teams. And here is what I’ve learned: accounting doesn’t need autonomy. It needs agency.
By autonomy, I mean agents that act independently. Agents that make decisions, execute workflows, and produce outputs without human oversight or a traceable chain of accountability. Accounting and auditing operate in a high-stakes environment, under strict regulations with zero tolerance for error. These industries don’t need “free agents” running loose.
The Work Is the Evidence
When I started Trullion in 2019, the thesis was simple. The PDF shouldn’t sit outside the workflow. It should be embedded inside it. Lease accounting and revenue recognition were never just calculation problems. There were documentation problems. The source documents mattered. The reasoning mattered. The audit trail mattered.
Thousands of corporations and accounting firms later, one principle has remained constant. In accounting, the work is the evidence. A number is only as strong as its trail. Calculations need to connect back to a contract, a ledger entry, a memo, or a judgment.
AI agents, as the tech giants are building them, optimize for answers. Accounting optimizes for defensibility. Those are fundamentally different goals. That gap is why the tools built for general-purpose AI fall short of what the profession actually requires.
The market today is fragmenting into three camps:
- Autonomous agents producing confident conclusions, often detached from embedded evidence.
- Workflow tools and spreadsheets organizing tasks, but disconnected from source documents and reasoning (no matter how turbo-charged by AI they are).
- Knowledge bases centralizing FASB, PCAOB, SOX, and policy guidance, yet floating separately from the work itself.
Yes, each solves a slice. None of them unifies evidence, workflow, and judgment into one continuous, traceable system.
The Horizontal AI Problem
What’s happening right now is something we’ve seen before. When a powerful horizontal technology platform emerges, the instinct is to push it into every vertical. Google was going to replace travel, retail, finance, and legal. It hasn’t happened. The verticals that demand deep specialization have survived and thrived.
The AI giants are running a version of the same playbook. Anthropic and Microsoft agents promise to automate “specialized” tasks. If you look closely at what they’re actually offering, you’ll see a collection of autonomous, disconnected agents. Each handles a single task in isolation, with no unified audit trail, no shared context, and no built-in accounting logic from the ground up. It’s fragmentation dressed up as innovation.
In a profession where every decision needs to be traceable and defensible, handing workflows to autonomous agents that operate independently is a new risk category.
Big Tech’s Pitch Doesn’t Add Up
I’d be lying if I said the foundation models aren’t really good. They’re getting better every month, which is why we continue to build on top of them.
There’s a difference between a powerful general-purpose engine and a finished product that an audit partner can stake their reputation on. Think of a large language model as a rare earth mineral. While it’s an incredibly valuable raw material, you wouldn’t hand a lithium deposit to a driver and call it a car. You refine it, engineer it into a battery, and put it in a vehicle designed for a specific purpose.
Similarly, accounting has a very specific purpose. When an auditor signs off on a financial statement, they are putting their firm’s name behind the numbers. When a CFO submits a regulatory filing, they are personally liable for its accuracy. There is no “close enough.” The tolerance for error is functionally zero.
Who Will Audit the AI?
Susan Parker, an accounting professor at Santa Clara University, recently told me that even with AI, she still believes students should review financial statements with a red pencil and physical tick marks. Why? Because tracing the statements manually forces them to feel the structure and nuance. It builds instinct.
It’s like learning guitar. AI can write good songs. Great musicians learn the notes, practice the scales, and understand how chords resolve. That effort builds mastery.
If young accountants only supervise autonomous systems, who develops the judgment to become the partner capable of making the hardest calls? Who audits the AI?
Professions are living systems. They evolve through practice. Abstract away the practice, and you erode the profession itself. Companies pushing autonomous agents in accounting don’t have that perspective.
What Auditable AI Actually Looks Like
The future of AI in accounting is what I call Auditable AI. It has three characteristics that the horizontal platforms, by their very nature, cannot replicate.
- Evidence is embedded in the workflow. Every output traces back to its source. When the AI processes a lease modification, you can see the contract it read, the standard it applied, and the logic it followed. The trail is built into the architecture.
- Workflows are connected to standards and domain knowledge. You can’t paste an audit methodology into a general-purpose chatbot and get a compliant workflow. AI needs to understand how practitioners tick and tie, navigate FASB and PCAOB standards, and how they connect firm-specific policies to document-heavy workflows. That domain expertise has to be baked in, not bolted on.
- AI is visible, inspectable, and interruptible. While the AI provides intelligence, speed, and traceability, the human makes the judgment. In a profession where liability is personal, you can’t outsource the responsibility.
This is what Auditable AI looks like in practice, and it’s what Trullion has been building since day one.
At the center of our platform is Trulli, Trullion’s domain-specific AI agent, grounded by Knowledge Room, a structured repository of accounting standards, firm policies, audit methodologies, and source documents. Unlike general-purpose agents that operate in isolation, Trulli works across connected workflows, allowing evidence, reasoning, and output to move together through a single continuous thread. Every action Trulli takes is traceable by design.
The AI revolution in accounting is real. Billions are being spent on powerful autonomous AI agents that promise to automate tasks across industries. In accounting, powerful AI requires auditability. The more capable these systems become, the more important it is that their work can be traced, verified, and defended. Companies building the next generation of accounting will be the ones that understand, from the beginning, that this profession runs on trust.
Accountants need agency. Not agents. In this field, the work is the evidence. The future belongs to those who can prove it.
