AI for Accountants Vol. 7 - Agents
Welcome to issue #016 of New Age Accounting — Vol. #7 of AI for Accountants. From mystery to employee. Here's what agents actually are — and how to deploy them responsibly.
When I heard the word “agent” a few years ago my mind went straight to Men in Black. Mysterious. Powerful. Operating in ways you couldn’t fully understand or explain.
That’s exactly how most accountants think about AI agents today.
That’s about to change.
What is an agent — actually?
Not the technical definition. The accounting definition. An agent is an automation of tasks and actions. AI has made them dramatically easier to create and deploy than anything that came before. Before AI, building an automation required developers, time, and significant technical knowledge. Now you describe the task, set the parameters, and the agent runs.
The analogy that actually works: an agent is an employee. It takes a task, executes it, and produces an output. The difference is it doesn’t sleep, doesn’t need onboarding, and can process data at a scale no human could match.
But like any employee — it needs oversight. It needs review. It needs someone who understands what it’s doing and can catch it when it gets something wrong.
How agents differ from everything you’ve learned so far
Quick orientation for readers who’ve followed the series:
Chat — you ask, Claude responds. One conversation, one output.
Cowork — Claude executes a multi-step task within a single conversation.
Projects — persistent context that carries across conversations.
MCP — the data connection that lets Claude pull live information from your systems.
Agents are what happens when you combine all of that and let it run autonomously. You’re not in the conversation anymore. You’ve defined the task, set it in motion, and the agent works through it — pulling data, performing analysis, producing outputs — without you watching every step.
That’s the power. That’s also the responsibility.
What agents actually do in accounting
Real use cases — not theoretical:
Pulling large datasets from your ERP and performing period over period analysis automatically.
Running headcount allocations and producing a journal entry upload ready for review.
Monitoring transactions against defined parameters and flagging anomalies without being asked.
Generating flux commentary after pulling current and prior period balances via MCP.
The pattern across all of them: the agent does the work. The accountant reviews the output. That human in the loop is non-negotiable.
Don’t treat them like a black box
Most people deploy agents the way they used to use Excel — set it up, trust the output, move on. That works until it doesn’t. And when it doesn’t in a financial context, the consequences are real.
You need to understand how your agent operates. What data is it pulling? What logic is it applying? What could go wrong and how would you know? These are the same questions an auditor would ask. Ask them before you deploy — not after something breaks.
The accountants who build agents responsibly aren’t the ones who know the most about AI. They’re the ones who think about their builds the way they think about their controls. With the end in mind. With an auditor in the room before the auditor arrives. (The Auditability of Agents)
The human in the loop
An agent without human review isn’t an employee. It’s a liability.
Every agent workflow needs a defined review point. Not “I’ll check it if something looks off” — a structured moment where a human confirms the output before it moves forward. This is what separates a workflow you can defend from one you can’t.
Think about it like this. You wouldn’t let a new hire post journal entries without review. You wouldn’t skip the approval process because they seemed competent. The same standard applies to agents.
Same controls. Different operator.
How to build your first agent
Practical and immediately actionable. Start with one recurring workflow — something you do every month, something rule-based and high volume. Describe it to Claude. Build the agent around it. Define the review point. Test it once before you trust it.
The first build changes how you see everything else.
From mystery to employee
A few years ago I didn’t know what an agent was. Now I use them daily to pull datasets, run analysis, and produce outputs that used to take hours.
They’re not mysterious anymore. They’re not Men in Black. They’re employees — powerful ones — that do exactly what you tell them to do and need exactly the oversight you’d give anyone else with access to your financial data.
Build them right. Review their work. Know what’s inside the box.
The Accounting Engineer doesn’t just use agents. They design them.
Vol. 8 is next. Keep building.
Here’s my question to you:
Are agents employees or liabilities in your book? Drop it in the comments — I read every one.
The purpose of New Age Accounting is simple: to empower accountants — at every level — to become builders, not bookkeepers. Whether you’re a staff accountant, a controller, or a CFO, there’s something here for you. Some topics will be high level, others will come with step-by-step guides, and some will include the exact prompts and tools you need to start building today.
If you’ve made it this far, you’re already thinking differently about this profession. Subscribe and come build with us.
Great article. Definitely Employees Brock, and need to be treated as such… continual coaching, development and check ins.