Today's CPA Magazine reports what we already know in practice: AI is now doing the work, and accountants are becoming supervisors. But there is a critical difference between what the Texas story describes and what UK regulated firms actually face. In the US, this is a workforce transition problem. In the UK, it is a regulatory competence problem. The FRC's ISA UK 240 standards, the SRA's rules on competence, and the FCA's Consumer Duty PS22/9 all place non-delegable responsibility on the partner and the firm. You cannot outsource AI governance to 'digital seniors' if your digital seniors have never been trained to audit an AI decision chain, challenge a model output, or explain why a depreciation schedule was wrong because the algorithm hallucinated a tax treatment. US firms are hiring for a new role. UK firms need to redefine who is actually liable.
What the Texas story reveals is the industry-wide assumption that the hard work is automation — teaching machines to do reconciliations, expense categorisation, and basic variance analysis. That work is largely done. The real work, the work nobody is systematically preparing for, is human judgment operating under machine-generated information at scale. This is not new territory for regulated firms in law, insurance, or finance — we have lived with this since Luminance and Harvey changed document review. But accountancy has taken a different path. Practices bought point-tool AI solutions (accounts packages with bolted-on reconciliation engines, expense categorisation widgets powered by Azure OpenAI) and assumed the transition was plug-and-play. It is not. The Texas article is telling us the reckoning has arrived.
Trovix's view is direct: the problem is not the AI. It is the infrastructure for human decision-making in an AI-mediated environment. Most UK accountancy practices have no formal framework for how a partner reviews an AI output, what questions they ask, what they document, and how they evidence their independent judgment to a regulator. They have process maps for the old workflow. They have tooling for the new one. They do not have governance. Compare this to how firms in legal services handled the same transition three years ago. Law firms using RAG-based tools like Aria know what they are auditing: the training data, the retrieval logic, the prompt, the output, and the user decision. They have built workflows around that reality. Most accountancy practices are still assuming a partner can spot-check AI results the way they used to spot-check junior work — faster, cheaper, with no intermediate controls. That assumption breaks down at scale and fails under examination by the FRC or the ICO. Trovix Watch exists partly because firms kept realising too late that their AI adoption strategy collided with a regulatory change they missed. The same applies here: you cannot govern what you do not see.
Here is what a mid-market UK accountancy or financial services practice should do starting now. First: do not buy another AI tool until you have documented who in your firm is competent to review its outputs and on what basis. Second: commission an audit of your current AI workflows — which decisions are human-led, which are machine-led, which are hybrid. Third: for each hybrid workflow, build a decision log: what did the AI output, what did the human change, why, and what was the business or compliance reason. This is not extra work; it is evidence work. It is what the FRC expects you to have done under ISA UK 240 when you have delegated a control to a system. Fourth: train your oversight team properly. Not on how the tool works. On how to challenge its outputs and explain their reasoning. The Texas CPAs calling this the rise of the 'digital senior' are optimistic. In the UK, it is the rise of the auditor — and if you have not started building that capability, you are behind.