The conversation about AI in tax has been almost entirely about tools, software that answers questions, summarizes documents, and drafts letters faster than you can. That era is already ending.
What's replacing it is something meaningfully different: AI agents. And the distinction matters more than most practitioners realize.
Tools vs. Agents: What's the Difference?
An AI tool responds when you prompt it. You ask, it answers. You stay in control of every step.
An AI agent acts. You give it a goal, research this issue, review this return, respond to this notice, and it executes a sequence of decisions on its own. It doesn't wait to be asked at each step. That's the shift that's happening right now across the profession, and it's arriving faster than the ethics guidance designed to govern it.
Several major tax platforms have already moved in this direction. The pitch is compelling: more capacity, fewer bottlenecks, faster turnaround. For a lean firm managing a large client load, the appeal is obvious.
What You're Responsible For
Here's the part that doesn't make it into the marketing materials.
Your signature is still on the return. Your name is still on the engagement letter. Under Circular 230 and the AICPA's Statements on Standards for Tax Services, the responsibility for the accuracy and adequacy of tax advice runs to you, regardless of what technology produced the first draft, the research, or the recommendation.
AI agents can make errors. They can misread a document, apply the wrong standard, or fail to flag an edge case that a human would catch. When that happens inside an autonomous workflow, the mistake may not surface until it's already baked into a deliverable. A tool that waits for your input gives you a check at every step. An agent that operates between checkpoints doesn't.
How to Stay in Front of It
None of this means avoid agentic AI. The firms that figure this out early will have a real capacity advantage. But there are practical things to get right before you deploy:
Define your review gates. Identify the specific points in any AI-assisted workflow where a licensed practitioner reviews the output before it moves forward. Document those gates. Don't let the speed of the tool erode the review.
Vet the vendor's liability framework. Ask explicitly: what happens when the AI makes an error? What logging exists? Who owns the mistake? If the answer is vague, that's your answer (i.e., you’re responsible).
Update your engagement letters. Disclose that your firm uses AI-assisted processes. Language around technology use is not yet standardized, but silence is worse. Your professional liability carrier may already have an opinion on this.
Train before you deploy. Staff who use agentic tools without understanding how they fail are a liability. The failure modes of AI agents are different from the failure modes of staff, and you need to train for both.
The capacity gains from agentic AI are real. So is the accountability. The practitioners who will benefit most are those who understand that the two don't cancel each other out but coexist, and that managing them is now part of running a firm.
Dr. Christine Gervais is a licensed CPA, using her skills to help businesses grow and achieve their fullest potential. Christine has a Master’s degree in accounting from Southern New Hampshire University in addition to holding her CPA license for over a decade. Notably, Christine is a nationally recognized speaker providing education to other CPAs on how to best serve clients as well as instruction on a wide variety of topics for business owners on how to maximize success. Christine prides herself on the value she can bring to clients with her extensive tax knowledge and provides strategic, forward-thinking financial strategies to help clients grow. When not behind her desk, you can find Christine spending quality time with her daughter and stepson or tending to the family’s excessively loved farm animals.
Like what you're reading?
Subscribe to our FREE newsletter and we'll deliver content like this directly to your inbox.


