How AI Is Actually Changing Accounting Right Now

Nine out of ten accounting firms are already using AI in some capacity. That stat comes from Future Firm, and it lines up with what we’re seeing on the ground. But there’s a big gap between firms using AI and firms using it well. Here’s where things actually stand.

The Practical Stuff Firms Are Doing Today

The most common use case isn’t some sci-fi automation pipeline; it’s client communication. Drafting emails, summarizing long threads, and classifying documents. Boring? Sure. But pausing mid-task to respond to a client email is a real workflow killer, and AI solves that.

Here’s a simple setup that works: use Zapier or Make for the automation layer, and Claude (via API) for the drafting. Five steps:

1. Trigger — A client email hits your inbox. Filter by sender domain or a “Clients” label so you’re not burning API calls on newsletters.
2. Categorize — Send the subject and body to Claude: “Classify this email as: invoice question, deadline request, general question, or urgent.” That category shapes the tone of whatever comes next.
3. Pull context — This is the step people skip and then regret. The client’s name, current engagement status, and last two or three emails in the thread help ensure the draft sounds like you instead of generic AI. If your CRM has an API, connect it here.
4. Draft — Send everything to Claude: “Draft a professional but warm reply from an accounting firm. Be concise.” Keep the system prompt short.
5. Review — The draft goes to your email drafts folder. A Slack ping flags it. You edit and send. Nothing goes out automatically.

That last part isn’t optional. Think of this as a drafting tool, not an autopilot.

ChatGPT is still the dominant tool in accounting firms, with Copilot, Claude, and Gemini trailing behind. Most firms are gravitating toward tools with simple chat interfaces — which makes sense. Accountants aren’t software engineers, any they don’t need to be.

If you want your AI to sound like you, start with three to five writing samples from your own emails and use them to build a short style guide. Feed both to the AI with a prompt like: “Analyze these samples and rewrite this message to match my tone.” Then keep refining. AI voice drift is real; it slowly defaults back to generic if you don’t keep training it.

An Intriguing Use Case: AI as the CFO You Can’t Afford

Kyle Ray runs Geek Window Cleaning and doesn’t have a CFO. He also can’t afford one. So, he’s using AI instead.

“Basically my entire leadership team is AI,” he told Business Insider. “I can’t afford the salaries for those kinds of people. But I can pay 40 bucks a month for Claude and ChatGPT.”

He uses it to research tax strategies, figure out whether a new hire should be a W-2 employee or contractor, and stress-test his own thinking. He still works with a CPA—and he should. AI is good at explaining concepts and organizing information. Applying that to a nuanced situation with real stakes is still a human job.

His take on trusting it: “I wouldn’t trust it 110%. I think it’s a good idea to always have someone look it over.” We agree—that’s the right mindset.

The Concept You Actually Need to Understand

Here’s the distinction most people gloss over: generative AI vs. agentic AI.

Generative AI responds to prompts. You ask, it answers. ChatGPT, Claude, Midjourney—all generative. Incredibly useful, but totally passive. The tool waits for you every single time.

Agentic AI pursues goals. You give it an objective, and it figures out the steps, takes actions, checks its work, and keeps going. You don’t babysit it through every decision.

Concrete example: with generative AI, you paste in financial data and ask about cash flow. With agentic AI, you say “review our new client onboarding queue, flag high-risk clients, draft summaries, and drop them in the project board by Friday”—and it actually does all of that across multiple systems.

Agentic AI is built on top of generative AI. It’s the same underlying model, but with added layers: the ability to take actions in software, maintain memory across tasks, and chain steps together. It’s the difference between knowing how to cook and actually running the kitchen.

Most accounting firms are still fully in the generative phase: You act as the project manager, and the AI is the very fast, very smart assistant.

That’s shifting. When AI starts sending emails, updating records, and triggering workflows on its own, bad instructions stop being an internal problem and start being a client-facing one. The stakes go up.

Understand the difference now, before you hand over the keys.