When I started work back in 1978 there were hordes of women working in typing pools. You had to write out your memo/document take it up to the pool and wait until it was done. Too many changes and it had to be done again as there was only so much twink could do !
Similarly there were a bunch of people employed in “Records”. If you wanted a file you had to request it and someone would have to deliver it for you. Any new documents / memos had to be added to the pile.
almost as bad in insurance coy offices in the 60s.
Insce proposals (the paper record of each policy) were in folders (30-50 proposals in each).
If someone else had the folder "out" (working on one of the proposals for a claim, insce cover change, renewal etc), you just had to wait until they'd finished, before you could access the one you needed....
Ai in 2026 is far more interesting as a tool for ministers to check the work of bureaucrats- as a strict productivity enhancer it has proved marginal thus far. Replacing staff with AI often just loads the thinking on to customers rather than actually doing productive cognitive work. Government risks yet another expensive IT disaster if it’s not very circumspect about implementation. And no, I’m not a Luddite - just realistic about what the tool can actually accomplish. Perhaps by the time AI is implemented it will have vastly improved…mind you it would be difficult to be less productive than a government department I guess !
Granted private business can be inefficient too, it's just that the cost of failure doesn't tend to fall on taxpayers like it does in the public sector.
Excellent commentary thanks Roger. But the only organisations that didn't reduce FTEs are government, both central and local. The main difference being that private sector customers can change who they buy from if prices get too high through low productivity or poor management. Public sectors have a captive market, because few political parties have ever been able to force better value from bureaucrats.
When I started work back in 1978 there were hordes of women working in typing pools. You had to write out your memo/document take it up to the pool and wait until it was done. Too many changes and it had to be done again as there was only so much twink could do !
Similarly there were a bunch of people employed in “Records”. If you wanted a file you had to request it and someone would have to deliver it for you. Any new documents / memos had to be added to the pile.
Whatever happened to all those people ??
They were some of the first to get their heads around computers. They learned “word processing” and “data entry”.
almost as bad in insurance coy offices in the 60s.
Insce proposals (the paper record of each policy) were in folders (30-50 proposals in each).
If someone else had the folder "out" (working on one of the proposals for a claim, insce cover change, renewal etc), you just had to wait until they'd finished, before you could access the one you needed....
Excellent analysis Roger. Should be read by all MPs and public service heads.
Great read, completely accurate.
Ai in 2026 is far more interesting as a tool for ministers to check the work of bureaucrats- as a strict productivity enhancer it has proved marginal thus far. Replacing staff with AI often just loads the thinking on to customers rather than actually doing productive cognitive work. Government risks yet another expensive IT disaster if it’s not very circumspect about implementation. And no, I’m not a Luddite - just realistic about what the tool can actually accomplish. Perhaps by the time AI is implemented it will have vastly improved…mind you it would be difficult to be less productive than a government department I guess !
Your last sentence is rubbish, Guy. You show a prejudiced mindset.
Granted private business can be inefficient too, it's just that the cost of failure doesn't tend to fall on taxpayers like it does in the public sector.
Correct, but it falls on their customers who, by and large, are the same people.
The future is the only undiscovered country…. And we can only know and appreciate it when we get there…
Excellent commentary thanks Roger. But the only organisations that didn't reduce FTEs are government, both central and local. The main difference being that private sector customers can change who they buy from if prices get too high through low productivity or poor management. Public sectors have a captive market, because few political parties have ever been able to force better value from bureaucrats.
Dear Roger,
Could you unpack what you mean by “a back-office analyst whose work is absorbed by a language model”?
Your argument seems to assume that analytical work in government is substitutable by AI, rather than fundamentally judgment-based.
Which parts of policy work do you believe are genuinely automatable, as opposed to merely assisted?
When you talk about work being “absorbed” by AI, are you arguing:
- that AI will mainly augment analysts and increase their productivity,
- or that governments will be able to materially reduce analytical headcount without degrading the quality of state capability and advice?
And how do you think governments can avoid hollowing out the pipeline through which future senior advisors and institutional expertise are developed?
Thanks