The government wants AI to cut 8,700 jobs, but doesn’t know how to use AI

 Printed by The Post for Stuff, May 2026

The people ordering a transformation of our public service just admitted they are using the most powerful tools ever created to….edit speeches. At yesterdays pre-Budget announcement Ministers said AI will become a basic expectation across departments, which is obvious and sensible (to say otherwise is akin to telling the public service to keep using typewriters after the PC was created). But what they didn’t provide the media was concrete examples of exactly how this transformation will happen and what work it impacts over the next three years.  

Going by released OIA’s earlier this year, in which one government minister asked AI to make her sound smart and impressive, the government is expecting rapid transformation of its departments without having implemented AI impactfully in their own backyard. Since they can’t, let me explain what they actually mean.

The robots are not coming. There are no robots that can replace you. However, it is knowledge work that this headcount target of 8,700 jobs will be largely drawn from. Yes, some public sector work will compress, and this will initially be the process driven 'rules based' work, but really the compression occurs around knowledge work.

That economics/ business/ humanities degree you got? It’s knowledge. That twenty years you’ve spent learning how to draft policies? That’s knowledge.

Those policy documents you analyse and write? That’s knowledge work.
Those templates, comms plans, strategy documents, precis papers, briefings you write? Also knowledge work.

Briefings, standard comms copy and the first analysis of consultation submissions - in these categories the first competent draft now takes minutes, not days (if you know how to use the tools). The volume of this work is largely fixed, so when the time-per-unit collapses, the headcount attached to it falls. The private sector is already experiencing this, and it is the part the Government has right (I won’t pretend otherwise: I spent years running a communications agency, and a great deal of what my team was paid to do can now be done by a capable person with a well set up AI system).

Knowledge work will shrink, and what we will end up with is smaller, sharper teams with AI on the judgement work. An example; the drafting collapses while the necessary judgement (is this option politically survivable, will this withholding ground survive the Ombudsman, is the briefing wrong because you’ve watched this minister for three years) expands to fill the space the drafting used to take. The judgement layer was, until now, trained by the process work. The grad policy advisor learned to read political risk by drafting OIA responses.

There is risk here – we don’t want a leaner senior workforce – we want a leaner sharper workforce with a pipeline of talent. The reality is that if this is done properly, this new leaner sharper workforce will significantly outperform the way our public service currently operates. It will also produce the productivity levels past governments have strived for.

For the taxpayer, this is a good thing. For our public servants, it is an era of change that requires an entire mindset shift in the way that they work. Again, this is like moving from a typewriter to a PC. And this will only happen if we enable people to really learn how these tools work, how to leverage them as a second brain, and use this new knowledge to innovate the entire way our public service functions.

This is the prize for those willing to support transformation: not doing more of the same work with fewer people, but redesigning how the work is done so people are doing the high value, innovative work that actually improves our public service. As far as I can see, we have no plan for this yet.

Now consider what we already know about ministerial AI use, because it has been tested. When Newsroom asked every ministerial office under the Official Information Act how ministers use AI, the answer from most was that they don’t. Some ended up looking foolish, and most looked like they use it only for the occasional chicken soup recipe. Those who released prompts were actually releasing their thinking out loud, the sort of brainstorming a minister once would have done with a spotty political advisor int he back seat of a Crown Car (‘I want to sound smart’, or ‘don’t say a home is an investment’ etc). This is a grey area that needs clarity, because in order to experiment safely and actually learn how these tools work (as a second brain) our public servants need to know that brain isn’t going to be exposed.

The Government can’t mandate deep AI use across the public service without personal privacy for the logs of the experimentation they are demanding of their staff over the next few years.

Someone has to decide which knowledge work compresses, which expands and which disappears. Which process work gets automated, and how. Which work becomes more senior, and where the talent pipeline comes from to support it. And someone has to resolve whether a public servant can think in an AI without their thinking becoming an OIA release. None of that is in the announcement. What is in the announcement is a reduction of 8,700 people by a government that is using the most powerful tools in the world as a chatbot.

Emily Broadmore is a former political advisor to government ministers, literary editor and workforce transformation strategist.