Three things AI should never take from tax professionals
AI agents can now do much of the legwork in tax research. William Young FCPA GAICD, founder of Praxio AI, explains why the thinking, the accountability and the training of our people must stay with us.
Over the past year, I have presented to accountants and tax professionals on AI in accounting and tax practice. Through building Praxio AI - Tax Assistant, I have also spoken with practitioners about where AI helps, where it creates risk, and what firms need to do differently in 2026.
The question in the room has changed. It is no longer whether we should use AI, but how far we should let it go.
That question is sharper now that AI is moving from answering questions to doing work. An agent can take a client scenario, research it and return a draft while you are in a meeting. Your clients are using AI too, and many arrive with AI generated answers they want you to confirm.
My answer rests on three beliefs.
1. Humans control the critical thinking, decision making and creativity
AI is very good at producing a first draft, and it is far quicker to review a draft than to write from a blank page. But a draft is not advice. We frame the question, test the facts, weigh competing interpretations and decide.
This is why AI is more valuable to an expert than to a client. Your client can generate an answer; only you know how to question it.
Treat AI like a junior you supervise. "Review these financials and give me advice" produces generic output. A better prompt sets the context, gives the AI a role, states the task, supplies the facts and says what to focus on. Then review, challenge and iterate, because using AI is never one question and one answer. Our users typically refine a scenario three to five times before they are comfortable with the result.
2. We cannot outsource accountability, duty of care or human review
The cautionary tales are familiar: a lawyer disciplined for citing cases that did not exist, a Big Four firm refunding part of a government fee, a bank reversing an AI call centre rollout. None is a reason to avoid AI. Each is a reason to have proper controls and governance.
Responsible AI is already part of our professional and ethical standards, and Australia's AI Ethics Principles, the Voluntary AI Safety Standard, the National AI Centre's Guidance for AI Adoption and the Tax Practitioners Board's new guidance statement all point the same way. You may use AI, but you remain accountable for the advice.
Governance does not need to start with a 50 page policy. Start small and build progressively: keep a register of how AI is used, define what it may and may not be used for, set rules for client data, and refine them as your use grows.
Then, for each piece of work, ask six questions:
What is the task?
What happens if the output is wrong?
What information am I entering, and does it need to be redacted?
Have I checked the sources? Ask the AI to link every source it relied on, then open them.
Who reviews it? Higher risk work should go to a manager or partner, just as it does today.
What evidence am I keeping for the file?
The aim is to move from "AI gave me an answer" to "I know the answer is correct".
3. We must invest in our people, especially the next generation
This is the belief I feel most strongly about. AI only works well for people who have already mastered their trade. If juniors use it to skip the reasoning, we risk a generation of practitioners who can produce an answer but cannot defend it.
AI should accelerate learning, not replace it. Ask juniors to form a view before they prompt, review the AI's sources together, and treat the review conversation as the training moment. Mentoring has always been how tax professionals are made. That job is ours, not AI's.
Putting these beliefs into practice
General AI is now extremely capable, but it is built to answer almost anything, and you have to steer it through every step yourself. We built Praxio AI - Tax Assistant to follow that discipline consistently, every time.
Praxio AI – Tax Assistant is built to follow the workflow of a tax professional, consistently: identify the issue, research credible Australian tax sources, analyse the position, consider alternatives and risks, provide a recommendation, draft clearly and keep a review trail. It also uses controlled, multi-step AI workflows to strengthen research and drafting. A second-pass review process challenges the initial analysis, checks for gaps and alternative interpretations, and helps improve the draft before it reaches the practitioner. These AI review steps support, rather than replace, professional review and judgement.
You can email a tax scenario to Praxio AI, it completes the research in the background and lets you know when it is ready for your review. The agent does the legwork. You still do the thinking and sign off.
Where to start
For firms implementing AI in 2026, my suggestion is simple: start small, set clear rules, protect client information, review every output and improve as you go.
Firms will not benefit from AI because they use the most tools, but because they use the right tools with good habits and clear governance.
Watch our webinar with The Tax Institute:
https://youtu.be/9omlSS78cqA
Start a free trial of Praxio AI - Tax Assistant:
https://app.praxio-ai.com.au/signup