The workings
Every number on the main page, with its arithmetic and its source. This page is here so you can check me.
The main page asks for money and makes claims. This page shows how each number was built and where each fact comes from.
If a source comes with a caveat, the caveat is printed here too. And where a number was built rather than borrowed, the maths is laid out so you can redo it yourself.
If you find something wrong, tell us and we'll fix it: mark@tenpasttomorrow.com
The costings
The Office of AI: $8m to $12m a year. That covers 15 to 25 full-time staff at senior policy grades, plus operating costs. It's a coordination cost rather than a programme cost. The training for ministers and chief executives barely registers against that, and there's no published figure for training at that scale.
The public service capability programme: $10m to $20m over three years. The public service was 64,535 full-time-equivalent staff at 31 March 2026 (Public Service Commission), across around 40 departments and departmental agencies. The costing takes $150 to $300 per person over three years.
This is the one figure in the document built from a headcount and a per-head range rather than taken from a source, which is why the maths is printed and anyone can redo it. It's a fraction of the per-teacher cost because there's no classroom to cover while someone trains, and everyone learns on one platform with shared systems and common standards. The teacher figure includes release time, which is the expensive part.
One million adults through a free foundation module. The ask is just the sum of its parts, so add the column yourself and you'll land on the same number:
- Content design and build: $3m to $5m
- Platform and delivery: $2m to $4m
- Activation and marketing: $8m to $12m
- Community and in-person delivery for roughly 150,000 harder-to-reach adults, at $150 to $250 a head: $22m to $38m
- Independent evaluation: $2m
Delivery through Māori and Pasifika providers happens inside this budget rather than as a separate line, so it doesn't change the headline number.
Split-cost vouchers for 30,000 small businesses at $1,000 to $1,500 of government share each, on the model of the existing Digital Facilitation Scheme, where the business pays the rest.
Every teacher in the country, at full price. A teacher workforce of around 68,900, at $400 to $600 per teacher per year, covering professional learning plus release time. That's the full-price figure on purpose, with nothing shaved off for teachers who don't take it up.
For scale: the ask averages $28m to $41m a year against national professional learning and development spend of about $138m in 2024/25. The closest like-for-like is the literacy PLD line at $40.7m a year, which is about the same size as this ask once you average it out.
Consenting capability, monitoring and enforcement: $5m to $8m a year. Potentially revenue neutral or better, through the conditions attached to consents.
The total
| Line | Low ($m) | High ($m) |
|---|---|---|
| 1 · The Office of AI, $8m–$12m a year × 3 | 24 | 36 |
| 1 · Public service capability programme | 10 | 20 |
| 2 · Learn it | 37 | 61 |
| 3 · Use it | 30 | 45 |
| 4 · Teach it | 83 | 124 |
| 5 · Look after it, $5m–$8m a year × 3 | 15 | 24 |
| Total over three years | 199 | 310 |
Stated on the main page as roughly $200 million to $310 million. Against the $76 billion opportunity by 2038, the total is 0.26 to 0.41 per cent, so "well under half a per cent" holds at both ends of the range.
The education comparison, all four ways
The main page says the ask is, at the lower end, less than half of what Budget 2025 reprioritised within the education vote. Here's that claim worked all four ways, so you can see exactly which one it stands on.
| Basis | This ask | Education reprioritisation | Ratio |
|---|---|---|---|
| Three-year totals, low end | $199m / 3 yrs | $613.5m / 4 yrs | 32% |
| Three-year totals, high end | $310m / 3 yrs | $613.5m / 4 yrs | 51% |
| Annualised, low end · the basis used | $66.3m a year | $153.4m a year | 43% |
| Annualised, high end | $103.3m a year | $153.4m a year | 67% |
The "less than half" line is only ever claimed against the annualised low end, and the main page says "at the lower end" out loud so the basis is visible. At the high end it doesn't hold, and the document never claims it there.
The sources
Each claim on the main page, with where it comes from. If a source has caveats, they're printed right next to it.
The economy
| Claim | Source |
|---|---|
| AI could add $76 billion to GDP by 2038, about 15 per cent of GDP | MBIE's national AI Strategy, which sources it to Microsoft research (2024). The main page attributes it that way on purpose: it's a vendor's estimate that our own government strategy chose to lean on. |
| Productivity gains from AI depend on adoption and diffusion, alongside complementary investment in skills, digital infrastructure, data, management practices and organisation | OECD AI Papers No. 29, Miracle or Myth? Assessing the Macroeconomic Productivity Gains from Artificial Intelligence (Filippucci, Gal and Schief, November 2024) |
| How well people work with AI depends on the user's experience and the task | OECD AI Papers No. 39, The effects of generative AI on productivity, innovation and entrepreneurship (Calvino, Reijerink and Samek, June 2025) |
| Up to $70 million over seven years for AI research and commercialisation, reprioritised from elsewhere in the science system | MBIE and Beehive announcements of the New Zealand Institute for Advanced Technology. Excludes GST; reprioritised from Callaghan Innovation and contestable funds. |
Business adoption
| Claim | Source |
|---|---|
| The Prime Minister is "constantly underwhelmed" by business AI adoption, said at the China Business Summit | RNZ, June 2026. Exact wording confirmed. In the same remarks he said government should lead on adoption. |
| 87 per cent of senior business leaders report their organisation "using" AI in some form, up from 66 per cent in 2024; 16 per cent transforming a core aspect; 12 per cent rolled out business-wide | Datacom 2025 State of AI Index. Sample: 200 senior leaders (C-level, senior managers, IT managers) at NZ organisations with 100 or more employees, fieldwork by Curia over one week in July 2025. So this measures our larger organisations only. |
| 97 per cent of New Zealand businesses have fewer than 20 employees | Stats NZ business demography, February 2025 provisional, out of roughly 612,000 to 617,000 enterprises. |
Adult learning
| Claim | Source |
|---|---|
| Digital Boost outcomes: 23 per cent of participating businesses reported improved revenue, rising to 39 per cent among weekly users; 80 per cent now have a website; 79 per cent would recommend it; 89 per cent intended to keep using it | MBIE Digital Boost Evaluation Research, August 2022, corroborated by the OECD's Observatory of Public Sector Innovation. Caveats: the evaluation itself calls it "still relatively early days", revenue improvement is self-reported, and about 32,000 businesses were registered at that point. |
| About $57 million was set aside for Digital Boost across its life | Treasury estimates and the Budget 2021 Cabinet minute: $13.5m identifiable in 2020/21, plus $44m operating over two years in Budget 2021. These are appropriations, money approved in a Budget rather than money proven to have been spent, which is why the main page says "set aside" rather than "spent". |
| A workplace first aid certificate takes twelve hours | NZQA consent and moderation requirements for unit standards 6400, 6401 and 6402; Toitū te Waiora. The full certificate is 12 hours; the 6401/6402 pair alone is 8 hours minimum. |
| Canada's AI for All strategy, launched 4 June 2026, funds a national AI literacy initiative and targets business AI use rising from just over 12 per cent to 60 per cent by 2034 | Government of Canada. It also targets up to 90,000 AI-related jobs and placements by 2031. |
Schools and teachers
| Claim | Source |
|---|---|
| All the schools figures: 93 per cent of leaders using AI; around seven in ten leaders with no data privacy and security policy for teacher use; three in four schools handling AI unplanned; 19 per cent of primary and around four in ten secondary students admitting misuse on assessed work; more than half of teachers concerned about critical thinking; 53 per cent of leaders and 73 per cent of teachers never having viewed the Ministry guidance | Education Review Office, Ready or not: How are schools responding to Artificial Intelligence?, 30 July 2026. Methodology: almost 4,000 survey responses (1,587 students, 859 teachers, 472 leaders, 889 parents and whānau), 129 interview and focus-group participants, 8 case-study schools, data collected Term 1 2026, English-medium state and state-integrated schools, Year 5 and above. The "seven in ten" phrasing is ERO's own. |
| ERO recommends a national framework, teacher and leader capability through targeted professional learning, and exploring centrally provided AI tools | The same report, recommendations 1, 8 and 9 of nine. ERO's Chief Review Officer publicly backed "a consistent, deliberate, strategic framework for all schools, a centrally-provided one" on RNZ. |
| Primary teachers largely working this out on their own, on free tools, most wanting more training than they get | NZCER primary teacher survey. Used for one supporting line only: the sample was self-selected and leaned towards teachers already interested in AI, so it's treated as the optimistic picture. |
| A teacher workforce of around 68,900: roughly 38,300 primary and 30,600 secondary | Ministry of Education, Teacher Demand and Supply 2025. The source frames this as projected demand holding steady over three years, and the main page is worded that way. |
| National PLD spend of about $138m in 2024/25, including $40.7m on literacy and Te Reo Matatini; no unified PLD quality framework in New Zealand, unlike Australia and Singapore | ERO, Teaching our teachers, 2025. Just over two-thirds of schools fund half or more of their PLD from their own operational budgets. |
| The official plan puts AI into the curriculum slowly: the Years 0 to 10 Technology curriculum released in September 2026 names AI inside its Computer Science strand, required from 2027 for Years 9 and 10 and from 2029 for Years 0 to 8; new senior subjects roll out for Year 11 in 2028, Year 12 in 2029, Year 13 in 2030, with a specialist Year 13 AI subject only being investigated | Ministry of Education and NZQA announcements on the NCEA replacement; Ministry of Education, NZC Technology Phases 1 to 4 (Years 0 to 10), September 2026, which states the 1 January 2027 and 1 January 2029 start dates. |
| Budget 2025 reprioritised $613.5m within Vote Education over four years; Budget 2026 reprioritised a further $65m | Ministry of Education and Budget documents. |
| Australia's framework for generative AI in schools has been in place since Term 1 2024 | Australian Department of Education. The Framework for Generative AI in Schools was approved by education ministers on 5 October 2023, implemented from Term 1 2024, and its 2024 review was endorsed by ministers in June 2025. |
The public service
| Claim | Source |
|---|---|
| 65,000 public servants | Public Service Commission workforce data: 64,535 full-time-equivalent staff at 31 March 2026, across around 40 departments and departmental agencies. The wider public sector headcount of about 477,400 is deliberately not used: the programme covers the core public service. |
Look after it
| Claim | Source |
|---|---|
| The Māori Data Governance Model, Tuia te korowai o Hine-Raraunga, published 2023, and the Māori AI Governance Framework | Te Kāhui Raraunga, with the Data Iwi Leaders Group, endorsed by the Iwi Chairs Forum. This document cites them as the standard to build on. |
| Te Pā Tūwatawata, a Māori-owned data storage network, launched May 2026 | Te Kāhui Raraunga. |
| The Human Rights Commission report, AI and Digital Technologies: A Human Rights and Te Tiriti o Waitangi Approach, 7 August 2026 | Human Rights Commission. A statutory Crown entity, so the government must table a response in Parliament. |
| Datagrid at Makarewa: a reported NZ$3.5 billion; consented March 2026; 280 megawatts to start and up to about 360 at full build; first phase due 2028; the second largest electricity user after Tiwai; a groundwater consent ceiling above 600,000 litres a day, about 220 million litres a year; the operator states free air cooling, rainwater harvesting and no groundwater draw | RNZ, BusinessDesk and the consent documentation from Environment Southland, Southland District Council and Invercargill City Council. It says "a reported NZ$3.5 billion" because published figures differ between reports and the company publishes no total capital cost. The application was not publicly notified, assessed as no more than minor in effect; it was notified to and approved by Te Ao Mārama Inc for the Southland rūnanga and Te Rūnanga o Ngāi Tahu. |
| Hundreds filled a public meeting at the Invercargill Workingmen's Club, organised by the Southland Sustainable Resource Coalition | RNZ. |
| Australia set up an Office of AI inside the Prime Minister's department in July | Australian government announcement, 15 July 2026. It sits inside an existing department; no director, budget or staffing have been published yet. |
Who's already doing this
Three Māori specialists who reviewed the fifth pillar asked us to point at the groups already working in each pillar rather than write as if we were first. Fair call. This list will grow. If your group belongs here, email us.
Lead it. AI Forum NZ, the industry body running the national AI conversation. The Government Chief Digital Officer, whose public sector AI guidance already exists.
Learn it. Digital Equity Coalition Aotearoa, the coalition working on the digital divide. InternetNZ, on digital inclusion and community engagement. Digital Boost, MBIE's programme that this pillar is modelled on.
Use it. AI Forum NZ again, and MBIE's Digital Boost delivery model.
Teach it. The Education Review Office, whose report this pillar is built on. Digital Future Aotearoa, digital tech education for tamariki and rangatahi.
Look after it. Te Kāhui Raraunga and the Data Iwi Leaders Group, who wrote the frameworks this pillar asks the government to adopt. Te Pā Tūwatawata, the Māori-owned data storage network. Te Ao Matihiko, with Te Matarau, the Māori tech association, and Te Hapori Matihiko, the Māori tech community. The Human Rights Commission, whose August report backs the principle.
Why the fifth pillar changed its name. Until 3 September it was called "Own it". The Māori specialists who reviewed it told us you can't own a language, a maunga or a river, only your relationship to them, and that a name arrived at without that conversation would read as appropriation. They were clear they weren't speaking for all Māori, and clear that the proper process, a wānanga, hasn't happened yet. This is where the name got to after one honest conversation, recorded here so anyone can see how it came about.
What other countries are doing
| Claim | Source |
|---|---|
| Canada: AI for All establishes a National AI Literacy Initiative offering entry-level AI training for all Canadians, reaching a million post-secondary students and more than 3,000 educators, with worker upskilling alongside | Prime Minister of Canada, "Prime Minister Carney launches AI for All", 4 June 2026. The release sets out the literacy commitments and the jobs and GDP targets, but does not publish a dollar figure carved out for literacy alone, so none is quoted here. Press reporting of the overall package ranges between about C$2.3 billion and C$3.5 billion depending on what is counted as new money. |
| Australia: the National AI Plan points over A$460m of existing programmes at AI research, skills and adoption, and makes foundational AI training mandatory for every public servant. A separate Office of AI was set up in the Prime Minister's department in July 2026 | Australian Government, National AI Plan, released 2 December 2025, and the Office of AI announcement of 15 July 2026. The A$460m is existing programmes redirected rather than new appropriations, and it spans research and ecosystem funding as well as skills and adoption, which is why the copy says "points at" rather than claiming it is all skills money. The plan and the office are seven months apart and the office does not run the plan, so they are described separately. |
| Finland: set out in 2018 to teach one per cent of its population the basics through Elements of AI, built by the University of Helsinki with Reaktor, hit the target, then offered the course across the EU | University of Helsinki. The one per cent target (about 55,000 people) was an enrolment target and was met. Course completion rates were far lower than enrolments, as with most free online courses, so the claim here is deliberately about reach rather than completion. Finland offered the course to all EU member states during its 2019 EU presidency. |
| UAE: AI is a compulsory school subject from kindergarten to Year 13 in government schools, phased in from the 2025 school year, with around a thousand teachers assigned to deliver it | The National, 4 May 2025, for the curriculum announcement; Gulf News for the teacher deployment, which was reported later in 2025 rather than in the original announcement. The subject sits inside an existing computing and design course rather than adding class hours. |
| United States: an April 2025 executive order made AI education for school-age children a national priority, leaning on public-private partnerships, with apprenticeships through the Department of Labor | Executive order "Advancing Artificial Intelligence Education for American Youth", 23 April 2025. It directs agencies and sets up a task force rather than appropriating money, which is why no figure is quoted. |
| China: national guidelines put tiered AI teaching into primary and secondary schools from age six; Beijing requires at least eight class hours a year in every school from September 2025 | Chinese Ministry of Education guidelines, May 2025, reported by Global Times and Xinhua, plus the Beijing municipal requirement reported by the State Council information office, March 2025. An earlier draft of this page wrongly described the eight-hour minimum as a national floor. It is Beijing's, not the country's. The national guidelines set a tiered curriculum without a published hours floor. Implementation detail beyond the guidelines is hard to verify independently from New Zealand, so the claim is kept to what is actually published. |
The 2001 conference
| Claim | Source |
|---|---|
| A three-day conference in Auckland, 1 to 3 August 2001, around 450 delegates from business, academia, economics, the public service and politics, 37 speakers including international guests from Ireland, Israel, Singapore, Canada and the United States | Conference communiqué and contemporaneous reporting. It was co-chaired by the Prime Minister of the day and the vice-chancellor of the University of Auckland. This document deliberately does not name the government or party involved, because the proposal is meant to be adoptable by any of them. |
| It changed the national conversation, and the follow-through was patchy | Both halves of that are contested and this document says so on the page. The conference is credited with shifting the argument about moving beyond commodity exports, and several institutions trace their origins to it. It has also been criticised at length since, including by business commentators, for how little of the agenda was carried through. The honest reading is that convening worked and delivery did not, which is the reason this document attaches deadlines and named owners to the proposal. |
The scoreboard
What we're sending. One letter, identical for every party, with the two-pager folded inside. It goes out on Tuesday 15th September, by email in the same hour and by post to each leader's office at Parliament the same day. The letter will be published here in full, as sent, on send day.
Who gets it, and the rule. Every party with a seat in Parliament on the send date, plus every registered party averaging one per cent or more, unrounded, across the three most recent published public polls on that date. The list and the three polls used will be named here on send day.
When. Sent Tuesday 15th September. A reminder on Thursday 1st October. Replies asked for by 5pm on Thursday 8th October, the day nominations close.
How replies are published. The part a party marks for publication goes up exactly as received. We correct nothing, not even a typo. Anything a party marks as background is not published. A pillar is marked as taken only when the party says so in its own words. Nothing on the scoreboard is inferred from what a party says anywhere else.
That's the workings. If you've read this far, you're exactly the kind of person this document was written for.