New Zealand’s Innovation Illusion- Stuck at Stage 1 AI While Its Innovation Engine Dies

As leading economies advance toward persistent agents and autonomous digital workers, New Zealand’s collapsing innovation throughput leaves it politically dense, strategically absent—and buying frontier technology without building frontier capability

As New Zealand heads into another general election and as ai-accelerates the world into the exponential age—and as we await the 2026 Global Innovation Index—the country risks another campaign of innovation slogans, isolated funding announcements and superficial technology promises. Before the next headline ranking arrives, the evidence already available demands a far more serious reckoning.

Christopher Luxon has said his government “inherited a mess.”

Yep, the innovation data supports that diagnosis- much of the deterioration was established before National took office to repair a very bad situation. But inheritance explains the starting position; it does not excuse strategic absence. Nearly three years into government, the question is no longer what National inherited, but whether it understood the structural failure and developed a credible strategy to reverse it.

Thus, New Zealand’s innovation problem looks manageable only when viewed from the outside. In 2017, the country ranked 21st in the Global Innovation Index with layers showing improvement. By 2025, it ranked 26th. A five-place decline sounds disappointing rather than disastrous. But that headline conceals what has happened inside the system. Across the measures connecting education, digital capability, research, knowledge creation and economic impact, New Zealand’s innovation engine has been progressively dismantled.

The critical question is not whether New Zealand retains sound institutions, publishes strategies or spends some money on research. It is whether the country can convert education-transformation (to Triple Loop, not entrenchment of single-loop) into expertise, expertise into intellectual property, intellectual property into scalable firms, and technological adoption into productivity and exportable value.

New Zealand’s Innovation Illusion- Stuck at Stage 1 AI While Its Innovation Engine Dies

That conversion process—innovation throughput—is where the rapid deterioration is unmistakable. It is also where the political silence is most damning, we say.

Innovation-system measure	2017 rank	2025 rank	Movement	What has deteriorated
Overall innovation	21 improving	26 rapid decline	↓ 5	The visible headline decline
Innovation inputs	13	22	↓ 9	Weakening capability entering the system
Human capital and research	17	23	↓ 6	Erosion of the national capability base
Education	13	33	↓ 20	A former strength becoming mediocrity; entrenchment of mass-delivery churn
Tertiary education	7	28	↓ 21	Collapse from near-frontier status
ICT capability	6	29	↓ 23	Loss of a crucial digital foundation
ICT use/Digital application	11	56	↓ 45	Technology availability failing to become deep use
ICT capability	6	29	↓ 23	Loss of a crucial digital foundation
Market sophistication	8	24	↓ 16	A weaker environment for financing and scaling innovation
Knowledge workers	29	37	↓ 8	A thinner base of people able to operationalise innovation
Knowledge and technology outputs	29	41	↓ 12	Poorer conversion into economically useful knowledge
Patents by origin	17	50	↓ 33	A severe decline in formalised intellectual property
Scientific and technical articles	7	15	↓ 8	Research production losing relative ground
Creative outputs	16	29	↓ 13	Deterioration beyond science and engineering
Knowledge impact	27	90	↓ 63	Near-collapse in the economic effect of knowledge
Labour-productivity growth	59	97	↓ 38	Innovation failing to improve national performance

Rank comparisons use New Zealand’s country profiles in the 2017 Global Innovation Index and 2025 Global Innovation Index. A lower numerical ranking is better.

Where it is recognised WIPO changes some indicators and data sources between editions, so no single movement should be treated as a perfect time series, this is not one anomalous indicator. Education, tertiary capability, ICT use, patenting, knowledge output, knowledge impact, productivity and creative output have all moved in the wrong direction. The breadth and internal consistency of NZ's decline make it impossible to dismiss as a methodological artefact.

Nor was 2017 a golden age. New Zealand already had weaknesses in science and engineering graduates, advanced manufacturing and high-technology exports. But it still possessed top-tier tertiary participation, ICT infrastructure, market sophistication, patent performance, research production and creative capacity. Those were the foundations from which a small, remote economy could plausibly build a knowledge-intensive future.

As de-globalisation has set in, since then, the country has retained much of the outer shell while allowing the internal machinery to corrode. Institutions remain comparatively strong: New Zealand moved from second to sixth.

Government digital services remain respectable. Business sophistication and some innovation-linkage measures have improved. Research expenditure increased modestly from approximately 1.2% to 1.46% of GDP, although this remains well below the intensity found in serious innovation economies. These fragments of strength make the failure worse, not better. New Zealand is not attempting innovation from a position of institutional chaos. It has stability, rule of law, functioning public administration and internationally credible universities. Yet it ranks 90th for knowledge impact and 97th for labour-productivity growth. That is a conversion failure of extraordinary proportions.

The education figures though, expose the beginning of the blockage. With industrial revolution practice entrenched in an era gone by with digital substitutes, at best, education fell from 13th to 33rd and tertiary education from seventh to 28th. The remaining school-level results are also less reassuring than they appear. In PISA 2022, only 71% of New Zealand students achieved baseline mathematical proficiency, compared with 92% in Singapore. Ten per cent were top mathematical performers; Singapore achieved 41% and Korea 23%. The OECD also cautioned that New Zealand’s results may have been biased upwards by approximately ten points because of low response rates and as it all is mass-delivery, chalk n talk, kids in rows, churn and still post-learning assessment. OECD PISA 2022 country note

The resulting pipeline is brutally clear-

Education: 13th → 33rd.Tertiary education: 7th → 28th.Knowledge and technology outputs: 29th → 41st.Knowledge impact: 27th → 90th.Productivity growth: 59th → 97th.

This is what a path to demise looks like—not the disappearance of every innovative company, researcher or entrepreneur, but the fundamental breakdown of the national system that should carry their work into global and regional economic scale at pace and sophistication.

Why this matters for Rare's AI framework

Our Rare AI capability framework makes the consequences immediate. Stage 1 AI—generative assistants and conversational systems—can be purchased from overseas and deployed without a sophisticated national innovation system. That is where much of

New Zealand currently sits - fiddling with licence-first, token toe in water pilots devoid of use-cases, chatbots, policy paralysis and workshops and scattered productivity improvements.

The Rare AI capability ladder

AI state	Indicative horizon	Defining capability	Principal wins and use cases	Strategic implication
Foundation- Narrow AI	Established	Prediction, classification and optimisation within fixed boundaries	Forecasting, fraud detection, diagnostics, scheduling and recommendations	Improves individual processes but does not redesign the organisation
Stage 1- Generative AI	Current mainstream	Produces and interprets language, code, images and knowledge	Drafting, analysis, coding, search, customer service and decision support	Rapid productivity gains; easily purchased and easily copied
Stage 1.5- Reasoning AI	Current–2028	Conducts deeper research, planning and multi-step problem-solving	Strategic analysis, clinical support, research, modelling and complex casework	Begins shifting AI from assistant to cognitive collaborator
Stage 2- Persistent agentic AI	Emerging now–2030	Maintains context, uses tools and completes workflows across time	Case management, procurement, finance, operations and autonomous coordination	Requires trusted data, system integration, permissions and redesigned work
Stage 3- Autonomous digital workers	2027–2035	Performs whole roles or operating processes with limited supervision	Digital teams, continuous operations, service orchestration and autonomous enterprises	Reshapes employment, management, organisational scale and competitive advantage
Embodied AI	Emerging–2035+	Connects machine intelligence with physical sensing and action	Robotics, logistics, manufacturing, agriculture, care and defence	Extends AI-driven disruption from knowledge work into the physical economy
Stage 4- Self-modelling systems	Uncertain	Models its own performance, limitations, goals and adaptation	Advanced autonomous research, scientific discovery and complex system governance	Creates major control, alignment, legal and sovereignty questions
Functional AGI	Uncertain; potentially 2030s+	Performs across most economically valuable cognitive domains	Cross-domain invention, management, research and problem-solving	Could compress decades of institutional and economic development
Stage 5- Synthetic agency/personhood	Speculative	Exhibits persistent identity, preferences and socially consequential agency	Autonomous representation, relationships and potentially independent economic activity	Forces unresolved questions of rights, liability and human status
Artificial superintelligence	No credible timetable	Exceeds leading human capability across virtually every cognitive domain	Scientific acceleration, system design and civilisation-scale optimisation	Potentially transformative—and existential—concentration of power

But Stage 2 persistent agents and Stage 3 autonomous digital workers require precisely the capabilities that have weakened: advanced education, deep ICT use, applied research, interoperable data, skilled knowledge workers, organisational redesign, intellectual-property creation and the ability to diffuse technology across firms and sectors.

A country ranked 56th for ICT use, 50th for patents, 41st for knowledge and technology outputs, 90th for knowledge impact and 97th for productivity growth is not approaching agentic AI from a position of strength. It is approaching it as a technologically dependent customer.

New Zealand can buy frontier models. It cannot thereby buy national capability.

The required response is a triple-loop national strategy. The first loop must deploy AI to improve present work and productivity. The second must redesign roles, workflows, institutions and education around effective human–AI collaboration. The third must anticipate and shape the emerging economy—building sovereign capability, locally owned intellectual property, new industries and export positions before foreign platforms determine the available choices.

At the centre of this should be AI-worker empowerment. The objective cannot be limited to automating jobs or reducing labour costs. It must equip New Zealand workers with secure AI agents, trusted data, advanced skills and greater capacity to analyse, decide, create and deliver. Properly approached, AI can make a small workforce disproportionately capable. Poorly approached, it will simply allow foreign technology owners to extract more value from that workforce.

Without a modern strategy model, and approach to linking education, compute, research, public data, worker capability, firm adoption, capital formation and export scale, AI will reproduce the existing pattern- foreign technology will be consumed locally, productivity benefits will be uneven, intellectual property will accumulate offshore, and domestic firms will remain users rather than architects of the emerging economy - for the exponential age.

This is why “politically dense, strategically absent” is not merely an insult. It is a description of the operating model. New Zealand produces strategies, working groups, ethical principles, announcements and pilot programmes. What it no longer reliably produces is throughput.

Triple-loop strategy would force government beyond announcements- from adopting tools, to redesigning the system, to deliberately shaping New Zealand’s position in the future economy. AI-worker empowerment would turn that strategy into national capability rather than another Wellington document.

The country has preserved the appearance of an innovation system while losing much of its economic and talent development function - in holistic terms. In an age of accelerating agentic AI, that is not stagnation. It is managed decline—and unless New Zealand changes course, it will become a tenant in somebody else’s intelligence economy (pun intended).