Everyone In

New Zealand has a productivity problem that has been stuck for a generation.

I'm making the argument that AI is the biggest productivity lever we'll see in our working lifetimes, and probably for a couple of generations after us. It's here now, and it's ours if we choose it.

A lever only works if people can reach it though.

The OECD has already made this argument for us. Its work on the macroeconomic productivity gains from AI finds that the payoff depends on how widely the technology spreads through an economy and on the complementary investment a country makes in skills and capability alongside it. That conclusion is the foundation of this document. What follows is what complementary investment could look like for a country our size.

New Zealanders are being told to use a technology they have not yet been enabled to learn. Fixing that is part economic policy and part fairness policy at the same time, and it will shape our working lives, our families and our children's futures more than anything else this generation will experience.

Somebody will ask why the government should pay for any of this when businesses already have every reason to get good at AI on their own. My answer to that is that the market reaches some people and stops. It reaches the firms big enough to have someone whose job AI is. It isn't reaching the sole trader, or the ward supervisor, or the 58-year-old who's just been told their role is changing. Nobody is incentivised to teach those people AI, so nobody does.

One thing to say before the questions. I work in AI. I'm not an education expert, an energy expert or a Tiriti expert. Where this document goes into those areas I've read the evidence and talked with people who do that work as domain experts for a living, and their fingerprints are on this. The AI parts are my thinking.

This document is not owned by anyone. That's purposefully the point. Any party, group or members of New Zealand society may take any part of it, put their name on it, and call it their own. I am writing this to encourage a conversation that I think is urgent, overdue, and that will meaningfully benefit New Zealand and all New Zealanders.

More details on me, and on the specialists I leaned on for research and review across the five pillars, are toward the end of this document.

Someone has to own AI in our country. Right now nobody does cleanly.

MBIE runs an AI economic strategy. The Government Chief Digital Officer runs public sector guidance on AI. Whether ordinary New Zealanders can use this technology belongs to neither of them wholly though, and so opportunities are slipping between the gaps. That's why the answer to "what is New Zealand doing about AI" is currently a lightweight national strategy.

Done properly this would mean less bureaucracy rather than more. MBIE and the GCDO should carry on with the work they are already doing well. What would change is that the same question would stop being asked in several places at once, and anyone trying to get an answer will have one door to knock on.

Australia set up an Office of AI inside the Prime Minister's department in July, pulling education, work, natural resources, energy, copyright and security into one place. We are asking for the same for New Zealanders.

And let's build a review point into it from the start. Five years in, somebody independent should look at whether the office is still needed in this shape, and say so publicly. Bodies that outlive their purpose are a fair concern, and we should plan for that.

And every minister and every public service chief executive should do twenty hours of AI training. A societal change of this consequence cannot be led from briefings. Our politicians must get their hands on the tools to tangibly understand them, so they can truly understand their power, their risks, and what they mean for all of us.

Twenty hours is under three days, and not much more than the twelve a workplace first aid certificate takes. It is no more than we would ask of a board member signing off a decision of this size. Set against everything else in this document it would cost almost nothing, and there is no shortage of very good people in this country who could run this training for our ministers.

Then the rest of the public service. The public service is one of the largest employers in the country. Its people assess the applications, run the payments, answer the phones and write the advice ministers act on. If AI is going to lift national productivity, and I truly believe it is one of the biggest levers we have, a large slice of that lift will happen inside government itself.

So this should be its own programme with its own budget, separate from the adult AI literacy programme in Pillar 2. About 65,000 people across around 40 departments and departmental agencies, on shared systems, delivered on-demand, with participation expected rather than optional. At $150 to $300 a head over three years that is $10 million to $20 million. Per person it costs a fraction of what reaching a million individuals costs in the Pillar 2 programme, because there is no classroom to backfill and the systems and standards are already built and common.

This is government going first, top to bottom, before it asks the rest of the country to do the same.

The AI Office will also work to the standard set out in Pillar 5 for how public sector AI handles Māori data. Consistency has to start here or it won't happen at all.

We keep telling New Zealanders to get on with AI. The Prime Minister himself has said he is "constantly underwhelmed" by how little our businesses are adopting AI. He is not wrong.

But look at what we are actually asking of people. To learn the most powerful consumer technology that's ever been created, that evolves and makes itself obsolete every few months. And to learn AI in your own time, on top of your job, with no guidance on what is safe, what is useful and what is nonsense. Then get told off for being slow.

We have done the alternative before, and it worked. In 2020 the government started Digital Boost, free online training that helped small businesses get to grips with digital tools. Across its life about $57 million was set aside for it, which is roughly the size of what this pillar is asking for.

And we know it worked because it was checkable and measurable. MBIE's independent evaluation in 2022 found 23 per cent of the businesses involved reporting improved revenue, rising to 39 per cent among those engaging with it weekly. 80 per cent had a website by then, 79 per cent would recommend it and 89 per cent intended to keep using it. The revenue figures came from the businesses themselves, so I'd call it a programme that worked, rather than proof of a productivity miracle. It's still the best evidence we have that a free, publicly funded capability programme changes what happens in a business.

Free foundation AI literacy for any adult who wants it. Online for most people, and in person for the people online will never reach, through the community and education channels they already use and already trust. A few hours to progress from AI nervous to AI capable, and a way to go further for those who want to.

It should be open to everyone, and budgeted honestly. The offer would go to every adult in the country. The costing assumes about a million of them will take it up over three years, because that is what meaningful and realistic take-up should look like.

And the providers should be funded based on results. Whoever delivers this training should be measured on whether people finish, and whether they can do meaningful things with AI afterwards that they couldn't do before. Evaluated independently, published, and funded against those results. It should be a national programme that is checkable, measurable and results-driven.

There's a second reason capability matters more than access. The OECD's review of the experimental evidence finds that what people get out of AI depends heavily on the skill and experience of the person using it. Handing someone a licence only goes so far. Teaching them how to use AI properly is what makes the difference.

As an example we can look to Canada, who started theirs in June. Their AI for All strategy funds a national AI literacy initiative alongside a target to lift business AI use from just over 12 per cent to 60 per cent by 2034.

Last year our government committed up to $70 million over seven years to AI research and commercialisation, money it found by reprioritising from elsewhere in the science system. That is a good investment and I support it. It also shows the money can be found when a government decides something matters.

Nothing comparable has yet been committed to whether the people who live here can use any of what that investment develops.

On jobs, because everyone is thinking about them. Some AI replaces tasks, and some AI makes the person doing the tasks better at them. Which of those a country ends up with depends partly on how much its people know about AI. Most roles will go through a period of profound change over the next few years, rather than instantly disappear. Then, in some sectors, real displacement will follow. We need to be honest about that. A country that has already built AI capability will be better placed to handle the second part. A country that has not, will not. Workplace redeployment support belongs inside this programme from Day 1, not bolted on later when AI's serious effects on the job market hit.

Most businesses in this country are small. Think a handful of people, no IT department, no innovation budget, and owners who are also doing the payroll on a Sunday night.

That is also the reason we could move faster than most nations. In a small business a good idea can go from someone noticing it one day to being how the business works three days later. In a large multinational the same idea will spend four months going through committees. Being small and flat is an advantage, but advantages have a shelf life.

On paper we are doing fine. Datacom's State of AI Index asked 200 senior business leaders at organisations with 100 or more staff, and found 87 per cent of them are "using" AI in some form, up from 66 per cent in 2024. Look closer though, and "using AI" includes businesses who have simply bought AI licences. Only 16 per cent say they are using AI to transform a core aspect of how they work, and 12 per cent have rolled it out across the whole business.

So, we ask for two things.

Split-cost vouchers, where government pays a share and the business pays the rest, so a small business can get proper help getting started with AI rather than watching YouTube videos at 11 pm at night. As mentioned above, we have run this before. Digital Boost paid $20,000 for each group of fifteen small businesses to work through digital change together. The wheel has been built before and MBIE knows how to roll it.

97 per cent of New Zealand businesses have fewer than twenty employees, and they're who this is for. Anyone calling it a gift to business should look at who is actually in the queue.

Training gets people using AI. Changing how the work is organised is what turns that into money. The businesses that get real value out of this will be the ones who redesign processes, rather than bolting AI onto the ones they already have.

And the second part costs the government nothing at all. The owners and leaders of our businesses learn AI themselves. A demo does not count. A lunch and learn briefing from whoever is good with computers in their company does not count. They need real hands on the tools training so they can feel the true power of AI. You cannot make good decisions about something you have only ever watched someone else use.

Then we need a measurement number of our own, for New Zealand. Not another adoption figure, because I have just talked about what adoption figures are worth. We need an AI capability target, and here's what I think it should mean. A business can name a meaningful process AI has changed, show you what it looked like before and after, and tell you who is accountable for it. Three questions, and you can answer them in a minute. You can't pass them by buying a licence. Let's set the target, publish it, and let us all be held to it.

In July the Education Review Office published the fullest picture we have yet of what AI is doing inside New Zealand schools. Nearly 4,000 people responded: 1,587 students, 859 teachers, 472 school leaders and 889 parents and whānau, alongside interviews, focus groups and visits to eight case study schools. Primary and secondary, Year 5 and up, gathered in the first term of this year.

It found that 93 per cent of school leaders are using AI. It also found around seven in ten of them have no policy covering data privacy and security for teachers using AI.

So, the people holding the most sensitive information about our children and their families are using these tools, and most of their schools have no policy covering the data that goes into them.

Three in four schools are handling AI as it comes at them, rather than with any plan. One in five primary students and around two in five secondary students say they have used AI on assessed work in ways they knew they should not. More than half of teachers are worried about what unplanned use is doing to children's ability to think for themselves. So am I.

We have given schools guidance. The Ministry of Education has a generative AI page for schools, put up at the end of 2024 and updated again this May, and NZQA requires every school assessing NCEA to hold an authenticity policy setting out where AI can and can't be used. 53 per cent of school leaders and 73 per cent of teachers have never looked at it.

Smaller New Zealand research from NZCER backs the teacher picture up: primary teachers largely working this out on their own, on free tools, most of them wanting more training than they are getting. That was a sample of teachers already interested in AI, which makes those the optimistic numbers.

Here's the part that should settle the argument. ERO went further than describing the problem. It made recommendations, and they have gone to the minister. Recommendation 1 is a national AI in education framework. Recommendation 8 is building teacher and leader capability through targeted professional learning. Recommendation 9 is exploring centrally provided AI tools for schools. The government's own review office is asking for three of the things this pillar asks for.

Meanwhile the official plan at the moment will only reach senior students in 2028 at the earliest, through the new qualifications rolling out for Year 11 in 2028, Year 12 in 2029 and Year 13 in 2030. A specialist Year 13 AI subject is still only being "looked into". For primary and intermediate there is nothing at all.

So we're asking for funding for every teacher to learn AI properly. By that I mean they will understand how AI can fit responsibly into education, building on the relationship between a teacher and a student, not replacing it. And they need release time so it happens in their work hours rather than on the couch at 9 pm at night. The Ministry of Education projects demand of around 68,900 teachers across our schools over the next three years. Done properly that would cost between $83 million and $124 million over three years, which is $400 to $600 per teacher per year. The assumptions are at the back.

That sounds like a lot until you put it next to what we already spend. New Zealand funded around $138 million of professional learning across the school workforce in 2024/25, of which $40.7 million went to literacy and Te Reo Matatini. Our ask averages about $28 to $41 million a year. One professional learning priority sitting alongside the others (not replacing them), at a comparable size to the literacy line, for the biggest change to impact their classrooms in generations.

ERO also found we have no unified quality framework for professional learning in this country, unlike Australia and Singapore, which means boards choose programmes with little assurance they work. So hold AI training to an evidence standard from day one. If we're asking for the money, we should be happy to be measured on it.

And no, I'm not saying this comes before literacy and numeracy. The basics must come first. The AI programme should phase in behind those basics, and it will protect them. Because if a teacher can't tell whether a child understands something or a machine did the understanding for them, we'll lose the ability to measure whether the basics are being learned at all. Getting AI right in schools is part of how we hold on to that.

Which is why assessment belongs inside the teacher training rather than beside it. Good AI training will teach teachers how to check for real understanding, at a time when a tool can imitate what learning looks like on the page. Universities are already working on this. The University of Sydney runs a two lane approach, where some assessment is supervised and secured and the rest is opened up to AI deliberately. That's worth our teachers looking at. We'll still need national guidance, but the work that matters most here is teacher capability, rather than a template.

Before any of that, the thing people are most worried about.

There are real fears that AI will wear away our children's ability to think for themselves, and that they'll stop trusting their own brains and their own instincts. I share them. Used well, AI can extend how far a person can think. That only works for someone who learned to think on their own first.

So the fundamentals still come first. We need an education system that keeps turning out strong thinkers, children who can reason a problem through, weigh up what's in front of them, and back their own judgement.

None of that is an argument against teaching our children about AI. It's an argument about the order we do it in.

Then a compulsory AI strand in the curriculum, built to be age-appropriate.

In the primary years, up to and including Year 6, children should learn about AI lightly, without really touching it themselves. What AI is, why it gets things wrong, and how to work a problem through on your own first. This is a suggested teaching sequence rather than a ban. The order matters, because the thinking habits have to come first, and be built carefully and intentionally in our children. That becomes more, not less important in the AI age.

Then from Year 7, we recommend safe AI tutor access for every student, so a child whose parents can't afford a tutor will get the same help as a child whose parents can.

And from Year 7 the teaching should cover more than which buttons to press. Children should learn where AI helps them, where it has no place in their learning at all, and when the right move is to close the laptop and think. Knowing when not to reach for AI is as much a skill as knowing how to use it, and it's the one that will still matter in twenty years.

One exception. All of the above covers general purpose AI tools. It doesn't cover assistive technology. Text to speech, speech to text, predictive text and reading support are built on AI now, and for a lot of children they are the reason school works at all. Those must stay available at any age, whenever a learning need calls for them.

A lot of how this goes will come down to parents. A big part of it will come down to teachers who have been trained well enough to teach it, which is where this pillar started.

Australia's education ministers approved a national framework for generative AI in schools in October 2023, and it was in classrooms from the start of the 2024 school year. They reviewed it and re-endorsed it in June 2025. We have Ministry guidance and nothing agreed nationally, which is exactly what ERO's first recommendation is asking for. Meanwhile it gets worked out here one staffroom at a time, and that is not fair on our teachers.

Left to chance, we see a probable pattern developing. Somebody else, meaning the overseas AI companies, builds the AI, we supply what it runs on, and the value ends up somewhere other than NZ.

Call it extraction.

It applies to our power and our water, and it also applies just as much to our language, our knowledge and our data.

Māori data and cultural knowledge is not raw material.

If a model learns te reo Māori, it will learn it from somewhere. Recordings, transcripts, waiata, whakapapa, iwi records, stories held by particular whānau. The same goes for mātauranga Māori. If those things get used by AI companies and AI models, the terms must be set by the people they belong to, and the benefit must come back to them.

None of that needs inventing, because the work is already done. Te Kāhui Raraunga, with the Data Iwi Leaders Group, published the Māori Data Governance Model, Tuia te korowai o Hine-Raraunga, in 2023, and the Iwi Chairs Forum endorsed it. They have since published the Māori AI Governance Framework, which extends that model to AI and treats Māori data as taonga.

So the ask here is a short one. The government should adopt those as the standard for how the public sector's AI handles Māori data, rather than commissioning its own framework and arriving somewhere similar three years later.

It's being built as well as asked for. Te Pā Tūwatawata, a Māori owned data storage network, launched in May this year. That's data held here, under Māori control.

There's independent backing for the principle too. The Human Rights Commission released a report on 7 August, AI and Digital Technologies: A Human Rights and Te Tiriti o Waitangi Approach, which puts Māori data sovereignty and Te Tiriti at the heart of how AI should get developed and governed here. The Commission is a statutory Crown entity, so the responsible minister has to table a response in Parliament. Dayle Takitimu, who leads its indigenous rights work, described what's needed as Māori being active partners in shaping Aotearoa's digital future. The Commission also made a valuable point: we don't have to start from scratch on any of this.

Underneath all of it sit four plain questions that get asked far too rarely. Where does the data physically live? Who can get at it? Whose laws apply to it? And who decides, at each stage of its life, what happens to it next?

Three practical things follow from this, and they'll be felt across all four of the pillars above.

In the adult AI literacy programme (Pillar 2), part of the delivery must run through the education channels Māori and Pasifika communities already use and already trust, instead of one mainstream pipeline.

In the small business programme (Pillar 3), we'll need clear expectations on consent, reuse and benefit sharing whenever a business points AI at Māori data, te reo Māori or mātauranga Māori.

In schools (Pillar 4), the AI strand must cover who owns data and who gets to say yes to its use, and schools will get guidance on how Māori student data is handled inside AI systems.

Now the physical resources side, where the same idea relates to power and water.

Earlier this year a data centre at Makarewa, just north of Invercargill, was consented. A reported NZ$3.5 billion, 280 megawatts to begin with and up to about 360 at full build, first phase due in 2028. Once it's running it will be the second largest user of electricity in the country, behind the Tiwai Point smelter. It has significant offshore backing, and the data it holds will leave the country down the trans-Tasman cable.

Here's the part that should stop us. The application was assessed as having no more than minor environmental effects, so it was never publicly notified. The country's second largest user of electricity was consented without the public being formally asked.

People turned up anyway though. Hundreds of them filled a public meeting at the Invercargill Workingmen's Club, with people left standing out in the hallway.

The consent lets them take more than 600,000 litres of groundwater a day, which is about 220 million litres a year. The company says it won't touch it. They'll cool the building with Southland air and collect rainwater instead, and I've no reason to doubt them. But the consent doesn't say that. It allows the water either way, it lasts for years, and nobody asked the public before it was signed.

(To be fair, Mana whenua were consulted on this one. The application was notified to and approved by Te Ao Mārama Inc on behalf of the Southland rūnanga and Te Rūnanga o Ngāi Tahu. My argument is with the framework rather than with this consent.)

I want data centres built here in NZ. They'll bring investment, they'll get built somewhere, and I'd rather it was us. This is not a call to stop them, or to pause for a year while we work out what we think.

Australia has shown the principle can be written down. Build a data centre there now and you must underwrite your own new power supply, including renewables, and local households don't have to carry that cost. You must pay for your grid connection and any water infrastructure you need, and you keep your water use down. None of it will stop data centres being built. But it does settle who pays the bills.

So we're asking for a national interest test before any data centre at this scale is consented, with the criteria published, and with iwi and the local community at the table rather than reading about it afterwards. Then let's put conditions on operators, along the lines Australia has set. How that gets written into law is the government's business.

Before this went anywhere near the public, I asked people who know many of these areas far better than I do to try to pull it apart. A NZ education specialist refined the schools pillar. An economist at one of our leading universities attacked my productivity argument the whole document is built on. A Māori data researcher examined the themes in the fifth pillar.

They all changed what I'd written, and meaningfully so.

This document was written by Mark Laurence. No party, agency, company or lobby group asked for it or had any hand in it, and it was written on my own time.

And yes, I run an AI training business, so I have a commercial interest in some of what this document asks for. I'm not going to pretend otherwise, or pretend I wouldn't want to help deliver it. My whole professional and personal mission is to help NZ achieve what I've described in this document. But what I won't do is use this document to hand myself work. If any of what this document proposes gets funded, it should go to open, contestable process, and my company can compete for it and be judged on the same terms as everyone else, or not at all.