When the junior stops asking the partner, how do they learn?

The way a first-year accountant learns tax has always had the same shape. They get stuck. They ask someone senior. They remember the answer. Over hundreds of these loops, judgement starts to form. The junior does not just learn the rule. They learn how to recognise a question that looks straightforward but is not, how to spot the thing a client did not mention, how to tell when something is worth escalating.

That loop has started to change. A first-year can now pull a source-linked tax answer in seconds. They do not need to walk to the partner’s office. They do not need to wait for a reply to a Teams message. The AI tool returns something that reads like an answer, with citations attached, before the junior has had time to ask.

For partners and firm owners, this is a real shift. The time cost of being asked constantly through the day falls. The quality of the first draft a junior produces goes up. The hours of research work that used to anchor a junior’s first two years get compressed into minutes.

The question is what that leaves behind.

The old loop was slow by design. The time a junior spent stuck on a question was time spent building the mental model of how tax actually works in practice. The partner’s answer was useful. The hour of looking at the problem before the question got asked was more useful. The pattern-matching that develops across hundreds of those hours is what makes a senior a senior.

When the research step compresses from an hour to thirty seconds, that pattern-matching does not happen on its own. It has to be designed in.

That is the question firms will face in 2027. The AI will keep getting better at research. The citations will keep getting cleaner. The gap between what a first-year can produce on the page and what they can defend in a client meeting is widening.

A reader might assume the professional framework has not caught up to this. It has.

The TPB’s primary supervision guidance, TPB(GS) 53/2024, applies to any staff member providing tax agent services on behalf of a registered practitioner, with or without an AI tool in the mix. Code item 7 requires that the service be provided competently. The guidance is explicit that “merely checking a document prepared by an unskilled employee to determine whether the contents of the document seem reasonable does not demonstrate a sufficient degree of supervision and control. There must be substantial supervision.”

Translated into the AI research setting, the TPB has already said what the review step cannot look like. A spot check on the final document is not enough when the junior has leaned on an AI to produce it. The supervision has to engage with how the answer was reached, not just how it reads on the page.

The regulatory position is the floor. The harder question is what firms build on top of it.

Three questions sit in front of every firm with juniors in it right now.

What do juniors still need to learn the slow way? Not everything is now a research question an AI can answer. The judgement calls that build over years of exposure, the client-reading skills, the sense of when an answer is too clean, the taxation law that is better understood from working it through than from reading a summary. The firm has to decide what the slow learning still looks like.

What does a partner’s time get spent on instead? If the research step is cheap, the partner’s involvement moves. The hour that used to go on “here is how to look that up” now has to go somewhere. The firms that have thought this through have moved partner time to the review and judgement conversations that previously got crowded out.

What does the review step look like when the research step has moved? This is the question with the sharpest edges. A review that would catch an incorrect citation does not necessarily catch an answer that is right on its face but wrong in context. The signals partners used to rely on in a junior’s work are not the signals that catch AI-assisted errors.

The Firm’s AI in Practice Summit on 12 November includes a session that works through exactly these questions. Michael Jeffriess, Co-founder and CEO of TaxGuru, makes a clear call on where this leaves firms in 2027. He walks through a complex tax question on screen with the sources shown and includes one answer the tool got wrong. The session ends with what firms can put in place in the next 90 days.

For firms thinking about how juniors learn in a year that looks nothing like 2023, the session is the shortest path to a clear view.

Tax Research: When the junior stops asking the partner: What your team learns in 2027

Michael Jeffriess, Co-founder and CEO of TaxGuru, walks through how firm training changes when AI gives first-years instant source-linked answers, what juniors still need to learn the slow way and the review step partners cannot delegate to the tool.

Thursday 12 November 2026, 4:00pm AEDT

Register for the session

Part of The AI in Practice Summit, 11 and 12 November 2026. Free, online, every session recorded.