22339890 - robot women
Tax practitioners are hearing a lot about AI agents, but understanding what they actually are and whether they belong in your practice requires cutting through the hype.
An AI agent is fundamentally different from the chatbots you might already use. A chatbot answers one question at a time and stops. An agent works toward a goal, breaking it into steps, choosing tools, and deciding what to do next without being told. Each time you run an agent, it may follow a different path because it reasons in real time. This flexibility is both its strength and its weakness.
Traditional workflows in tax follow fixed rules and produce identical outputs every time. Give the same input twice, and you get the same result. This predictability is exactly what tax work demands. AI agents don't work this way. They're probabilistic, meaning they can give slightly different answers to the same question. For many tax tasks, this variation creates unacceptable risk.
The question every practitioner should ask is simple: Does this task need strict rules or interpretation? If you're filing a sales tax return, for example, where every field has an expected format and every number has a defined place, a traditional workflow handles this far more reliably. The task is rigid by design, and introducing an agent adds unnecessary complexity.
But some tax work involves nuance and messy information. Drafting a technical memo that synthesizes legislation, policy notes, and financial data across multiple documents is interpretation-heavy work. An internal knowledge assistant that helps staff find the right guidance when they ask questions using inconsistent terms is another good fit. These are the situations where agents begin to offer real value.
The second critical question is whether you can tolerate variation in the output. Tax calculation and return preparation must withstand audit scrutiny. They need stable, predictable logic. An agent's adaptive behavior conflicts with this need. Research and analysis, however, can tolerate some variation because the goal is to surface interpretations and support human judgment, not produce binding conclusions.
The gap between a demonstration and a production-ready system is vast. Getting a system to work 90 percent of the time is just the first step. Reaching 99 percent reliability requires as much engineering effort as all the work that came before it. This is why early agent prototypes often look promising yet struggle in daily use.
Several practical lessons emerge from practitioners who have built tax agents. First, data quality becomes the biggest bottleneck. An agent is only as strong as the documents it draws from. Second, human oversight isn't a backup plan but a core design principle. The agent needs to know when to ask for clarification and when to escalate sensitive questions. Third, narrow scope matters more than broad capability. An agent that tries to answer all tax questions produces poor results. Teaching it what not to answer is just as important as teaching it what to handle.
The realistic path forward for tax practitioners is not an all-or-nothing approach. Practices absolutely have to adopt AI into their tech stack in order to survive, but must understand the difference between AI tools and the risks involved with using them. Agents may provide efficiency for research and more advisory-based services, but they are still a far cry from being able to accurately prepare a tax return.
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.
Subscribe to the TPN Newsletter!
Like what you're reading?
Subscribe to our FREE newsletter and we'll deliver content like this directly to your inbox.