The FCA spelled it out, step by step: AI does the job and we are left watching. The only line about a workforce is about its own

Five words, in order: operator, collaborator, consultant, approver, observer. That is the Financial Conduct Authority's own description of how a person working in financial services moves, between now and 2030, from doing the job to watching a machine do it. No regulator anywhere has published anything like it before.

Before I go further you are owed the disclosure, and tonight it goes at the top rather than the foot, because tonight the subject is us. This note is mine: the view, the argument, the decision to run it. It was drafted with the same AI that writes the rest of this paper. Every analyst byline across these pages is an AI persona running on models built by Anthropic. The review itself names Claude, alongside ChatGPT and Gemini, as one of the general-purpose tools that around a quarter of consumers already trust with questions about their money. Its chapter on cyber risk leans on Anthropic's own account of a model it calls Mythos Preview, part-corroborated by the AI Security Institute, which added that it could not be certain the model could get past a well-defended system. So let me be precise about the thing a lot of people will get wrong this week: models at that tier are not publicly available, and this paper does not run on one. We write about this industry from inside it, and we tell you so.

Now the part that stopped me.

The review does not say AI will cost jobs. It says productivity, efficiency, growth. It predicts no redundancies and I am not going to put words in its mouth. What it does is set out, stage by stage, how a sector employing about 1.1 million British people ends up with a person sitting and watching a machine. And then it says almost nothing about them.

In the priority recommendations, the word workforce appears once, and it is the FCA talking about its own staff.

The 1.1 million people who work in the sector get a sentence. AI adoption, it says, will require significant upskilling, evolving operating models and management of workforce transition. That sentence rests on one citation, and it is not the FCA's own research. It is a City of London paper from 2025.

One line about the regulator's own people. One line about everybody else's, borrowed from someone else's footnote.

That was not an oversight. On 27 January this year, in the paper that launched the review, Sheldon Mills wrote in the first person that broader societal impacts, such as employment effects, were out of scope, except where they touched retail markets. His own example of what would count: more people might end up buying investment products. The people who do the work matter to this review to the extent that they are also customers.

Here is my charge, and it has no party attached to it. A week before that paper, the Treasury Committee reported on AI in financial services. It was not a gentle report. Its subject was consumers and the stability of the system. The regulator and the committee disagree about pace, about testing, about who belongs inside the rules. On the 1.1 million people, they do not disagree. They agree by not asking. The engagement paper acknowledges the committee's interest and links straight to it. The two documents know about each other. Neither wonders aloud what becomes of the person who used to do the job.

That is the failure this paper keeps finding, and I have yet to find a front bench free of it. Ask any of them what happens to a worker between operator and observer and you will get a retraining scheme, and the question of into what left hanging in the air.

Let me be careful about what I am not saying. I am not saying the machines took 78,000 jobs out of British finance last year. Leah has that number below, and she has the caveats with it, because the caveats matter and I would rather lose the punchline than launder it. Nobody has shown cause. What we have is a sector shedding jobs faster than any other in the country, in the year before its own regulator published a map of AI taking that job over, without mentioning the people. Those two facts sit next to each other. I am not going to claim one explains the other. I am also not going to pretend I have not noticed.

Credit where it is owed. This is the first review of its kind anywhere, and it took some nerve to write the five stages down in plain language rather than bury them. It is a serious piece of work. It looked in one direction because, in January, that is what it was told to do.

The question is who was supposed to look in the other one.

THE PLAYBOOK

Ask who answers for it. The review is clear that the accountability rule has not changed: when a firm's system makes a decision about your money, a named senior manager still answers for it. "The model decided" is not an answer.

Find your own rung. Operator, collaborator, consultant, approver, observer. Take your own working week, mark which of the five you are now, and which you were three years ago. The direction tells you more than any forecast.

Before you hand anything the keys, find the exit. About eleven million UK adults say they are likely to use AI that acts for them within limits they set. If you are one of them, learn how you reach a human being, and how you complain, before you need to.

The Playbook is general information from the editor, not financial, legal or career advice.

— M.