Could one person really do five jobs with AI?

I was reading about the Kashim Chowdhury case and it got me thinking.

Chowdhury is the civil servant who held several full-time public sector jobs at the same time. Between March and May 2023 he held five full-time jobs at once and was contracted for 183 hours a week. There are only 168 hours in a week. On 9 November a jury at Southwark Crown Court convicted him of nine counts of fraud (CPS ).

The question was simple. Could one person really do five jobs with AI?

The short answer is no. Not five, and not today. I went through the job specs, and in the best case AI might let one person cover about two of these roles. The people work, the meetings and the days in the office don’t shrink.

But the question doesn’t go away. And the more I looked, the more it led somewhere more interesting. We’re scared of AI taking our jobs. We also want to pay less tax and get better public services for the tax we do pay. We want it both ways. I think we can have it.

What AI can and can’t do in a public sector job

AI can already take on most of the writing and admin in a public sector project role. It can’t do the people work those roles are built around.

I wanted to know what these jobs actually involve, so I went and found the job specs. His employers included Defra, the UK Health Security Agency and Tower Hamlets Council. I matched the kind of role to real adverts, mostly project and programme management. They’re remarkably alike. Every one asks for plans, risk logs, budgets, reports and governance, delivered through other people. Each is full time, and most expect two or three days a week in the office.

Then I sorted the duties by how much AI could help with them.

• A lot: status and board reports, meeting notes and actions, plans, schedules, and risk logs. AI can draft these from the data, flag slippage and chase actions.
• Some: business cases, budgets, and performance measures. AI can structure, reconcile and analyse. A person still has to get the numbers agreed and argue the case.
• Very little: stakeholders, line management, presenting to boards and ministers, steering suppliers when delivery slips, and being in the room. AI can prepare you for these. It can’t do them for you.

That matches the wider research. The Alan Turing Institute found that 47% of civil servants’ working time involves tasks AI could help with. But only 4 of the 91 activities it studied could be fully handed over, and all four were admin (PublicTechnology ).

So here’s the maths. Take those five jobs and 183 contracted hours. Now assume AI does every single task it could help with, which no study claims. You’d still need 97 hours a week, more than two and a half full-time jobs. Plus up to 15 office days in a five-day week.

My estimate of today’s realistic gain is more like 2 to 9 hours a week per role. That’s time to do one job better. It isn’t room for a second.

What one person can now deliver with AI

One person working with AI can now get through a remarkable amount of work in a day. How much depends heavily on the kind of work.

We’ve seen this inside Fifty Five and Five. Last month my colleague Chris Wright logged everything he did in a single working day, working with a set of AI agents. By the end of it he had:

• Analysed 530 answers from ChatGPT, Claude, and Gemini to see how a client shows up in AI search
• Written 55 help pages and nine web pages for a product he’s building
• Found 17 new people to approach, drawn from past clients, job moves, and company news
• Rewritten titles, descriptions, and calls to action across more than 100 of our blog posts
• Shipped several software releases, and still got through his emails and meetings

I’ve worked in agencies for almost 30 years. When I started out, several of those would have been projects in their own right.

AI gains most on new builds like these. Work that runs through other people, which describes most public sector roles, gains less. And every piece of it still needs a person to decide it’s good enough. AI agents can review the work, often more thoroughly than we used to. A named person still signs it off.

We use AI on our own work first, so we can see where it helps and where it doesn’t (more in our AI adoption guide ). The honest picture is that it makes every kind of work faster, by very different amounts.

AI in the public sector: two headlines, one number

The same AI saving can be told as a story about lost jobs or a story about lower taxes. Both would be true.

I pushed my analysis to its best case. Assume AI does every task it could help with across government’s project delivery staff. By my estimate, that frees up about 12,000 posts’ worth of time, worth around £551m a year in salaries. It’s a ceiling, not a forecast. Now write the headline.

“12,000 public sector jobs lost to AI.”

“AI saves taxpayers £551m a year.”

Same number. Very different reactions. And the public holds both views at once.

On one side is fear. In a King’s College London survey, 57% of the UK public said they expect AI to cause widespread unemployment. Only 17% think it will create as many jobs as it takes, or more (King’s College London ). Only 7% think the gains will be shared fairly.

On the other side is the tax bill. The latest British Social Attitudes survey found a record 19% now want taxes and spending cut, more than double the 8% in 2022. Support for higher taxes and more spending has fallen to 36%, the lowest since 2013 (NatCen ). Most people want something simpler: better services for the tax they already pay.

I understand both reactions. If it’s your job, the first headline is the only one that matters. If you’re waiting for a hospital appointment, the second one looks pretty good.

The third headline

There’s a third way to tell it, and it’s the one I care about most. AI does the legwork so people can do the caring.

Think about what those freed-up hours could become. A public health manager who spends less time on commissioning paperwork and more time with residents. NHS managers who hand the savings tracker to AI and spend that time with clinical teams. A project team that gets its weekends back because the weekly report writes itself, and it’s up to date.

That version doesn’t happen by default, though. The savings only fund more human care if someone decides they should.

Why AI productivity numbers are probably too cautious

I think the numbers I’ve used are on the cautious side, and the savings will grow.

The best-known UK figure comes from the government’s own trial. More than 20,000 civil servants used Microsoft 365 Copilot for three months and reported saving 26 minutes a day, mostly on drafting, summarising, and updating records (GOV.UK ). It’s a useful number. But it was self-reported, it was published in June 2025, and the tools have moved on a long way since.

How far? Researchers at METR measure the length of task an AI system can complete on its own. They found that length has doubled roughly every seven months since 2019 (METR ). If that holds, the tools in that trial are already about three doublings behind.

There’s a fair challenge on the other side. A separate Department for Business and Trade trial of Copilot found no discernible productivity gain, and the Institute for Government has told MPs that savings on the scale of the government’s £45bn estimate would need “pretty meaningful headcount reductions” (The Register ).

Both things are true. Giving people a licence doesn’t change much on its own. The savings come when you change how the work is done. That matches what we see with clients, and it’s why the next part matters.

Make AI part of the job description

Could five jobs be done to a sufficient standard with AI? No, not yet. But for the jobs that already exist, using technology to do the job better should be part of the job description. Free up time, save costs, provide a better service.

Every spec I read asked for the same things: manage the budget, own the risk log, bring stakeholders with you. I didn’t see one that asked people to use AI to get there.

So what if they did? What if every public sector job spec asked for it, right alongside budgets and stakeholder management? What if the weekly report took minutes, and the time went into the work nobody ever has time for, like talking to residents or keeping suppliers on track? What if teams were judged on the service they provide, not the paperwork they produce?

We’ve tried a version of this ourselves. On an AI search project for a global software client, we started by doing the work by hand, month by month, the way an agency traditionally would. Once we’d proved the process worked, we built a tool to do it and replaced ourselves. Not the client’s staff. Ourselves.

That’s the shift. AI isn’t something to hide or to fear in a role. It’s part of doing the role well.

So, could one person really do five jobs with AI?

No. Not five, and not yet. AI can take on most of the writing and admin in a public sector role, but the people work, the presence, and the accountability don’t compress. In the best case, one person might cover about two of these roles.

That isn’t really the interesting question, though. The interesting one is what we do with the time AI frees up in the jobs that already exist.

We want it both ways. We don’t want AI taking jobs, and we do want lower taxes and better services. I think we can have both, but it won’t happen by accident. It needs four things:

1. Make using AI part of every job description. Not a side project. Part of doing the job well.
2. Judge teams on outcomes and service, not on the paperwork they produce.
3. Put the freed-up time and money into direct human care. More time with residents, patients, and the people the service is for.
4. Let roles change openly and gradually, not through sudden cuts.

Do that, and AI in the public sector stops being a story about lost jobs or a story about savings. It becomes a story about better services, delivered by people who finally have the time to deliver them.

We want it both ways. We can have it, if we choose to.

If you’re thinking about how AI could free up time in your own team, I’d love to hear about it. Drop me a line.

Frequently asked questions