Kia ora! Welcome to New Zealand’s weekly roundup of AI news, events, jobs and education.
This video caught our attention this week.
Isaac Asimov gave robots three rules: don’t hurt humans, obey humans, protect yourself. Unitree’s G1 Humanoid Robot Max apparently read them, then ignored them the second someone said “role-play”. Check the video here.
This week’s highlights:
Zespri ships 215M trays on AI
AI translates NZ news into te reo
86% of Kiwis back AI for 111 calls
Antioch raises $7M for robot sims
Beca cuts 4-year project to 12 months
Agentic commerce: the quiet B2B shift
OpenAI launches GPT-5.2 in three tiers
Anthropic hits 40% enterprise AI share
Ben Reid: NZ's tech sovereignty at risk
Happy reading and listening ✌️
Did someone forward you this? Sign up!
MIKE’S MUSINGS
Forget AI keynotes: AI Lab Job Boards Reveal Where AI Is Headed In 2026
OpenAI is becoming an infrastructure company, 450+ roles heavy on data centre technicians and robotics. Anthropic is playing trust, with 30% of hiring in sales and B2B for regulated industries. xAI ran up 100,000 GPUs in 122 days versus the typical four years. DeepMind is hiring a "Research Scientist, Post AGI Research".
Picking an AI provider is now a strategic bet on which future wins.
12 DAYS OF AI CHRISTMAS
The 12 Days of AI Christmas is a short daily series from The AI Corner NZ sharing 2026 predictions from 20+ Kiwi AI leaders.
Each day focuses on one theme so readers can spot patterns forming across the NZ ecosystem, not just chase tools. Sharing the first few days of predictions:
Day 1: Agentic AI & Workflows | How work is shifting from manual coordination to AI systems running end to end. That sets the foundation for everything else, because how work runs determines what skills actually matter.
Day 2: AI in Schools & Universities | Moving upstream into education and capability, with specialists inside schools and universities sharing what they are seeing on the ground. The clear message so far is that NZ’s biggest risk is not access to AI, it is whether we build the judgment and fluency to use it well.
Over the next ten days we will keep layering themes to show where AI in NZ is heading and what to pay attention to now. Follow along!
💼 Business & Industry
Kiwifruit packhouses are now running 24/7 on AI. Zespri will ship a record 215 million trays this season. Mount Pack & Cool went from 6 million to 30+ million trays in six years using automation that solves the $50/hour worker shortage.
4-min read.
Our take: When packhouses can't fill roles at $50/hour, the economics flip permanently. Early automation adopters are building cost advantages that late movers won't catch. The next generation of packhouse workers will troubleshoot AI systems, not stack boxes in the cold.
"Neanderthals" have 18 months before AI reshapes knowledge work. Patel predicts all knowledge work will be done with AI tools within 18 months. Nightingale counters that the profession has adapted before - from adding machines to internet. The industry's hourly rate model is dying. 40% of young workers now report AI-induced anxiety about their futures.
5-min read.
Our take: The real disruption isn't AI replacing accountants, it's AI destroying time-based billing. Firms selling hours instead of outcomes will be obsolete.
NZ's top law firm charges AI tools directly to clients. Russell McVeagh uses pay-per-transaction AI for e-discovery and due diligence. CEO Ben McLaren says the cost-cutting mindset around AI is wrong, it's about creating more value.
4-min read.
Our take: Passing AI costs to clients is the new business model. Law firms that absorb these costs will struggle against those who position AI as premium service delivery. Russell-Mc’s new CEO is a breath of fresh air in a traditional industry. Let’s hope the long awaited shift doesn’t mean they’re too far behind.
Beca claims AI compressed a $150M project from 4 years to 12 months. Digital twins let designers, builders, and owners collaborate in real-time. NZ faces a $275B infrastructure pipeline through 2055 that existing resources can't deliver.
3-min read.
Our take: Building the wrong thing in the wrong place is NZ's most expensive infrastructure mistake. Digital twins might finally break that cycle. The infrastructure deficit is so massive that AI isn't optional, it's mathematical necessity. There simply aren't enough humans to plan and build what we need.
NZ startup Antioch just raised $7M to simulate real-world robot testing. Former Tesla engineer Harry Melsop's software creates digital twins that compress months of testing into seconds. Antioch's software lets companies test agricultural robots virtually before deploying them physically.
7-min listen.
Our take: Testing robots costs a fortune because real-world environments are expensive to rent. Simulation-first development could slash those costs dramatically.
Year-long mainframe migrations now compressed to weeks using AI agents. Datacom says agentic AI can interrogate legacy systems, understand undocumented code, and suggest optimal migration paths. The shift from generative to agentic AI happened faster than expected.
3-min read.
Our take: Alongside modernising migrations, the real value here is knowledge recovery. Decades of institutional knowledge trapped in undocumented systems is suddenly accessible again.
Pie Funds adopts Marloo AI to kill admin bottleneck. Wellington startup Marloo raised $4.6M to automate post-meeting compliance documentation for financial advisers. Pie Funds formalised the partnership after trials showed significant efficiency gains.
3-min read.
Our take: Advisers spend hours on file notes instead of finding clients. Marloo just bought back the most valuable asset in financial services, time to actually advise.
⚖️ Society & Education
AI now translates NZ news into te reo Māori at speed of the news cycle. Stuff partnered with Straker and Microsoft to translate content at scale, reviewed by human translators. Output doubled from 4.3 to 8.9 articles per day.
3-min read.
Our take: One language dies globally every two weeks. Using AI to keep te reo thriving while maintaining quality through human review is exactly the right balance.
86% of Kiwis comfortable sharing exact location with emergency services. New research shows strong public support for AI detecting critical keywords like "knife" in emergency calls (58%) and ranking calls by urgency (55%). Current tech is 60+ years old.
3-min read.
Our take: The next generation of emergency callers expect media-rich interactions. Building for voice-only is building for the past. The 60-year-old systems we have are being outpaced by smartphone capabilities. The gap between what the public expects and what 111 can actually do is a clear market signal. Modernisation investment is both wanted and overdue.
📚️ AI for Business
Helping leaders and teams adapt, learn, and scale with AI.
1️⃣ Productivity is now about orchestration and not focus. The old model was grinding on a single task for hours. The new commercial reality is spinning up 15 agents to execute work in parallel. Real value now comes from planning and refinement rather than manual execution.
4-min watch.
2️⃣ Agentic commerce is the quiet B2B disruption. McKinsey values the US market alone at $1T by 2030 but the real story is B2B which is 4x to 6x larger than B2C. If AI agents cannot read your pricing and inventory via API then you are invisible to the modern buyer. B2B is the critical battleground here because agents thrive on structured purchasing data.
1-min read.
3️⃣ OpenAI data shows the gap between toys and tools. There is a massive divide between tinkerers and transformers. Users tackling seven task types save 5x more time than those using four. Frontier firms send 7x more messages to Custom GPTs because they are building institutional memory. You are either building power users or you are just paying for subscriptions.
2-min read.
4️⃣ The multiplier effect of AI on business fundamentals. Adoption is not about hype but it acts like leverage. Move fast on shaky foundations and you multiply your operational problems. Move fast with solid education and governance and you multiply your enterprise value.
2-min read.
5️⃣ Microsoft targets were missed due to business strategy gaps. The issue is not that agents make mistakes. The issue is that leadership teams do not understand what they are buying and data foundations are not ready. We are seeing a mismatch between vendor promises and enterprise readiness.
2-min read.
6️⃣ Efficiency is useless if you do not reinvest the capacity. Anthropic found that engineers increased output by 50% while AI usage jumped from 28% to 60% of their work time. Critically they found that 27% of the work would not have happened otherwise. The commercial question is where you redirect that freed capacity to generate net new revenue.
2-min read.
7️⃣ The $200 stack is a workforce alternative. Wispr Flow, Claude, and Perplexity cost roughly $200 NZD per month. Do not view this as a software cost. View it as the price of a digital workforce that delivers the output of a junior analyst for a fraction of the salary liability.
2-min read.
🎙️ The AI Corner Podcast
This week's guest is Ben Reid, Technology Futurist and Founder of Memia (pronounced "‘Meme - ee -ah’). Hear:
Why New Zealand has blindly surrendered technology sovereignty to a handful of US hyperscale cloud providers.
Ben's prediction of an imminent AI investment correction and what it means for building on cheaper infrastructure.
How open source investment is the only viable path forward for countries outside the US and China.
🌍 Tech Updates From Global
The selected top headlines from each major AI tech company.
OpenAI
Launched GPT-5.2 (codenamed "Garlic") on December 11 in three tiers: Instant, Thinking, and Pro, with improved coding, math, science, vision, and long-context reasoning capabilities.
Secured $1 billion equity investment from Disney alongside a three-year licensing deal making Disney the first major content partner on Sora, with 200+ Marvel, Pixar, and Star Wars characters available for AI-generated content from early 2026.
Released State of Enterprise AI Report showing ChatGPT now has 800 million weekly users and enterprise message volume grew 8x since November 2024.
Co-founded the Agentic AI Foundation (AAIF) with Anthropic and Block under the Linux Foundation, donating AGENTS.md to establish open standards for AI agents.
Anthropic
Announced multi-year partnership with Accenture forming the Accenture Anthropic Business Group, training approximately 30,000 professionals on Claude, one of the largest Claude practitioner ecosystems worldwide.
Co-founded the Agentic AI Foundation with OpenAI, donating its Model Context Protocol (MCP), now the standard for connecting AI models to tools with 10,000+ published servers.
Menlo Ventures report shows Anthropic now holds 40% enterprise AI market share and 54% of the AI coding market, up from 32% enterprise share in summer 2025.
Chief Scientist Jared Kaplan warned AI could do "most white-collar work" within 2-3 years and predicted humanity will face the "ultimate risk" decision on self-training AI models by 2027-2030.
Google/DeepMind
Launched reimagined Gemini Deep Research agent on December 11 alongside new Interactions API, enabling developers to embed research capabilities into their apps.
Announced sweeping UK government partnership including first automated research lab opening in 2026 focused on discovering superconductor materials and new semiconductors.
Selected by US Department of Defense's Chief Digital and AI Office to deploy Gemini for Government across 3 million civilian and military personnel via GenAI.mil.
Microsoft
Announced $17.5 billion investment in India over four years (2026-2029) (its largest-ever investment in Asia) for cloud and AI infrastructure, with commitment to train 20 million Indians in AI by 2030.
Committed $19 billion CAD ($14 billion USD) to Canada between 2023-2027, launching a new Threat Intelligence Hub in Ottawa and partnering with Cohere for sovereign AI models.
Announced strategic partnerships with Cognizant, Infosys, TCS, and Wipro to deploy over 200,000 Microsoft Copilot licenses collectively, setting new benchmark for enterprise AI adoption.
Amazon/AWS
Published re:Invent 2025 roundup featuring Kiro Autonomous Agent, an AI coding tool that maintains awareness across sessions and handles bug triage across multiple repositories, described as "orders of magnitude more efficient" than first-generation tools.
Announced general availability of Multimodal Retrieval for Bedrock Knowledge Bases enabling AI-powered search across text, images, audio, and video.
xAI
Announced partnership with El Salvador to deploy Grok AI across 5,000+ public schools, creating world's first nationwide AI-powered education program reaching over one million students.
Elon Musk confirmed Grok 4.20 release within 3-4 weeks (late December/early January), representing one of xAI's fastest upgrade cycles.
Integrated Grok into Tesla vehicle navigation, allowing AI to serve as a "personal guide" for adding and editing destinations through natural conversation.
Perplexity
Faced federal copyright lawsuit from The New York Times alleging large-scale unlawful copying of millions of articles, with NYT claiming Perplexity made over 175,000 access attempts to nytimes.com in August 2025 while ignoring robots.txt protocols.
Launched AI patent search beta, Email Assistant (with privacy guarantees), live flight status tracking, and expanded sports coverage across 10 new leagues.
📅 AI Events in New Zealand
The great wind-down begins. Only 6 AI events happening across the country this week.
This week’s featured event:
AI Transformation Beyond the Hype, today in Auckland. I’m hosting a panel with execs from Fonterra, Overdose, and academyEX to dig into what’s actually driving AI ROI.
📅 Promote your event with us. Reply to let us know.
💼 AI Roles Around Aotearoa
Picklist of 🌶️ HOT 🌶️ new roles in AI this week
GTM AI Specialist, Halter: Auckland
AI & Data Consultant, EY: Wellington
💼 Promote your job with us. Reply to let us know.
🤦 ️ AI Funny Of The Week
We all love AI, but it’s certainly far from perfect …
👋 Mike & Erin