If AI does the coding, what’s left for the bookkeeper to check?

AI can now handle much of the capture, coding and review in a bookkeeping file. That frees up time, and what a bookkeeper does with it is starting to matter more than the hours it saves.

Some bookkeepers are putting that time into the judgement calls their clients value most. They pick up the transaction that looks wrong. They answer a client’s question before it is asked. Over time, they become someone the business would struggle to run without.

For a registered BAS agent, those judgement calls sit inside obligations the Tax Agent Services Act 2009 already sets. AI taking over the mechanical steps leaves those obligations where they were, with the agent.

Section 90-10 of the TASA defines a BAS service as a tax agent service that relates to “ascertaining liabilities, obligations or entitlements of an entity that arise, or could arise, under a BAS provision”. The definition also covers advising an entity about those liabilities and representing it in its dealings with the Commissioner.

There is a second condition. The service must be provided in circumstances where the client can reasonably be expected to rely on it to meet its obligations or claim its entitlements. The definition turns on what the service relates to and whether the client relies on it. How the service was produced is not part of the test.

Section 50-5(2) sets a civil penalty for providing a BAS service for a fee if you are not a “registered tax agent or BAS agent”. A business can use whatever software it likes to process its transactions. Paying someone to provide a BAS service is different, because the provider generally has to be registered. AI can do a lot of the work in a bookkeeping file, and the registration sits with the agent.

Section 30-10 of the TASA sets out the Code of Professional Conduct that registered agents work to. Because section 90-10 defines a BAS service as a type of tax agent service, the Code items written about tax agent services apply to BAS services too.

Code item 7, in section 30-10(7), deals with competence: “You must ensure that a tax agent service that you provide, or that is provided on your behalf, is provided competently.”

The obligation covers the whole service, including work done on the agent’s behalf. A BAS service produced with AI has to meet the same standard as one produced without it.

Code item 8, in section 30-10(8), deals with keeping up: “You must maintain knowledge and skills relevant to the tax agent services that you provide.”

The skills relevant to bookkeeping work shift as the work itself changes. Code item 8 makes keeping those skills current an ongoing obligation for every registered agent.

The first judgement call is spotting the transaction that looks wrong. A tool can code a transaction. Recognising that it looks wrong for this particular client takes an understanding of how that client’s business runs. When the coding feeds a BAS, getting that call right is part of providing the service competently under Code item 7.

The second is answering a question before the client has to ask it. That means reading the file closely enough to see an issue coming and raising it early. AI assistants inside bookkeeping tools can now flag some issues on their own. Deciding which ones matter to this client, and how to explain them, depends on the knowledge and skills Code item 8 requires agents to maintain.

The third is becoming someone the business relies on. That comes from doing the first two well, over time, for the same client. Reliance is already part of the legal definition. Under section 90-10, a BAS service is one the client can reasonably be expected to rely on to meet its obligations or claim its entitlements.

AI changes how much of the agent’s time goes on capture and coding. The obligations in Code items 7 and 8 stay with the agent either way.

The TASA sets the obligations. How a bookkeeper rebuilds their week around AI is a decision each practice makes for itself.

The Summit session on AI-assisted bookkeeping shows what that looks like right now. It covers where AI handles capture, coding and review and where a bookkeeper’s judgement matters most. It also looks at AI assistants that answer questions and flag issues before they reach the client, and at MCP servers that connect bookkeeping data to tools like ChatGPT and Claude.

Glenn Castle, Chartered Accountant and Regional Product Domain Expert at Dext, is joined by a tech-savvy bookkeeper who’s building their own AI-powered workflows. Together, they show what this looks like right now. The session covers where AI assistants take work off the team and where a person still needs to be in the loop. It also shows what MCP servers make possible when bookkeeping data can talk to tools like ChatGPT and Claude.

Thursday 12 November 2026, 11:00am AEDT

Register for the session

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