A fifth of London's jobs. That is the number the Greater London Authority put on paper in April 2026, and over a million people sit behind that statistic wondering what it means for them. Economist Jeff Dwan-O'Reilly's report is worth reading past the headline — not because it resolves the uncertainty, but because it maps it with unusual precision.
I build these systems. I know what they are capable of, and the GLA's numbers look different when you spend your days designing the pipelines the report is warning about.
Which Jobs Are Actually at Risk?
The report identifies more than 300,000 administrative roles as particularly vulnerable — the kind of work that lives inside email clients, spreadsheets, document editors, and CRM systems. These are exactly the environments a capable AI agent can operate in end-to-end, without a human in the loop. The blast radius of a well-configured AI agent in a standard London office is not a single conversation window: it is the entire software stack a junior administrator uses to get through their day.
A further 748,000 roles in IT support, data analysis, and secretarial work face significant exposure, though impact varies by specific task. That variation matters. A data analyst who spends 70% of their time pulling and cleaning data is in a very different position from one who spends 70% interpreting results and advising on strategy. The report does not always draw that line clearly — which is one of its limitations.
Who Bears the Cost?
The demographic skew is the detail most likely to surprise people: women, young workers, and those with higher educational qualifications are disproportionately at risk. That last point runs against the intuition that education is protective.
It is not, here. Generative AI is most capable at knowledge work — precisely the territory of graduates in administrative, analytical, and secretarial roles. A first-class degree followed by three years as a policy analyst at a mid-size consultancy does not insulate you from a model that can read, summarise, cross-reference, and draft at scale. If anything, it puts you closer to the target.
The 54% the Report Buries
The report presents its finding that 54% of London's workforce occupies roles with "limited AI exposure" as reassurance. I would read it differently.
Forty-six percent facing meaningful exposure is not a minority concern — it is close to half the workforce of one of the world's largest financial and professional services cities. The roles cited as low-risk — architecture, culinary arts, executive management — are either physically embedded (you cannot debone a fish remotely) or require accountability structures AI currently cannot satisfy (a board will not accept a model as a signatory). These are real constraints, but they are also narrow categories. "Limited exposure" does not mean permanently safe. It means not yet.
What the Report Actually Recommends
Mayor Sadiq Khan and Jeff Dwan-O'Reilly's report does not argue for slowing AI adoption. It argues that a laissez-faire approach to the transition — letting displacement happen without deliberate intervention — would cause serious, concentrated harm to London's labour market. The ask is proactive policy: retraining infrastructure, sector-level preparation, and coordination between employers and government on where exposure is highest.
That is a reasonable position. It is also a position that has been staked out at every major technological inflection point for the past two centuries. What is different this time is not the shape of the argument — it is the speed of the change and the breadth of roles affected simultaneously.
What This Means in Practice
The augmentation-versus-replacement framing in the report is useful but incomplete. Augmentation is real: a data analyst with a well-configured AI co-pilot can accomplish in a day what previously took a week. The question is whether that translates to one analyst doing the work of five, or five analysts doing the work of twenty-five. The answer determines whether the net employment effect is neutral, beneficial, or damaging — and nobody has that answer yet.
What I am confident of, from building these systems, is that the transition will not be gradual in the way policy tends to assume. It will be uneven. Some roles will hit a cliff edge the moment a capable model becomes available and affordable for a specific task. Others will be unaffected for years. The GLA report is a useful map of where London's workforce is most exposed. Using it well means acting before the cliff, not after.