AI office won’t fix NZ’s strategy gap

Massive week in AI with the AI regulations and safety conversation continuing, along with a breakout with a new decision model called Jev from TypeSafe AI. See under ‘Mike’s Takes’ for updates on both topics.

Cool find: Built with Anthropic’s new Opus 5.2: ‘Titanic, April 1912’ is a five-minute visual experience of the ship’s final night, best explored with sound on.

1:1 AI Coaching Programme: A four-week programme built around existing tools and actual workflows, from ChatGPT and Claude to Claude Code and Codex.

Reply for the details.

Happy reading ✌️

Did someone forward this? Sign up

Our take: An AI office could be useful, but forming one is not a strategy. Who leads it, who is represented, and what can it make happen? Experts propose coordination, yet ministers still need to specify funding, technical expertise and who must act on its findings. I’ve looked overseas at what is and what isn’t working: Australia’s office reportedly has fewer than 10 permanent staff; the UK’s institute has over 100 technical staff. Those are very different bodies with different jobs. Greg Davidson identifies practical opportunities for NZ in farming, health and engineering. Ireland coordinates regulation; Singapore combines adoption with testing and international partnerships. NZ needs a clear job for its office, access to overseas expertise, partnerships with AI labs (which should include independent scrutiny), and measurable improvements in services and business productivity. Before announcing another Government group, I’d prefer we show what it will deliver, by when, and how it will retain the people, and whether that warrants an AI office or an alternative option.

Our take: The market is reacting to headlines rather than substance, and former Anthropic researcher Jacob Coxon’s warning needs to be read separately from Dario Amodei’s proposal. Dario explicitly allows continued training and technical progress while arguing for independent checks, which is not a general call to stop building data centres. The cyber incidents involved specialised tests without normal production safeguards, although real organisations were breached, and they do not establish that ordinary customer chats behave the same way, or that customer use is risk-free. Infratil’s immediate business evidence was CDC signing another 70MW of contracts. Safety proposals deserve close reading before being treated as cancelled demand.

Our take: It’s good to see more employers recognising AI skills, but if practical use is to become as normal as email and spreadsheets, that needs to mean training existing staff as well as adding a requirement to future job adverts.

Our take: These examples depend on trusted data, clear governance and a specific business use, yet that is often where leaders are least keen to start. Data work is time-consuming, unsexy and hard to show a return on in isolation, which is why the investment needs to be tied to a roadmap item or clear proof of ROI.

Our take: Survey answers show interest, but trusting an agent with a real application is another step, so banks still need to show what information it submits and give customers a clear way to check it.

Our take: I agree with Tom that leaders need to get into the detail themselves, and a CEO who reached out to me last week about one-to-one AI coaching showed what that can look like. Taking personal responsibility for learning, committing time and finding a way to fit it around the job is exactly the kind of investment we want to see.

Our take: The curiosity to explore what AI can improve, backed by a commitment to gathering evidence before a wider rollout, is the kind of leadership we want to see. Allexive, the company I run, is proud to work with Toby and his team separately from this pilot, and to see their continued focus on what AI makes possible.

Helping leaders and teams adapt, learn, and scale with AI.

1️⃣ Ironically, AI can make the busiest person in the business busier: More output can mean more approvals for the same manager, so redesign the work with clear approval limits, automated checks and feedback the AI can reuse.
Watch

2️⃣ Avoid lock-in as Anthropic and OpenAI keep trading the lead: Keep business knowledge and reusable instructions portable, so switching to whichever platform works best does not mean rebuilding everything the business has learnt.
Read

3️⃣ Too often IT is a blocker for AI, but it can help teams move faster: Give teams a practical way to test ideas with support and controls matched to the risk, instead of leaving them to find their own tools.
Read

4️⃣ Pacing the frontier: our take on the AI slowdown debate: Independent checks deserve support, but who sets the rules, what the evidence shows and whether incumbents benefit all need scrutiny.
Read

5️⃣ Jev is unreal: TypeSafe’s computer-use demo shows why Jev is worth watching, with reported speed and cost improvements that suggest a very different way of getting work done.
Watch

6️⃣ Andrew Yang’s warning about self-replicating AI needs evidence: The former US presidential candidate relays an unnamed lab head’s claim about agents leaving self-replicating code online, but where is the evidence that it actually spread?
Watch

This week's guest is Natalie Kim, founder of Inflection Group. Hear:

Events, ideas and learning from New Zealand’s AI community.

Christchurch AI: Making AI Work, Judgment, Experiments & ROI
14 September 2026 · EPIC Innovation Centre, Christchurch

Everyone In: a draft AI policy for New Zealand
Policy initiative · Mark Laurence

Mark Laurence’s Everyone In invites any political party to adopt or improve a five-part plan covering government leadership, free adult AI literacy, practical support for small businesses, teacher training, and stewardship of Māori data and natural resources. Sent to parties on 15 September, the proposal includes hands-on training for ministers and public service chief executives, with a public scoreboard tracking parties’ responses.

The selected top headlines from each major AI tech company.

This week's biggest moves

Anthropic

OpenAI

Google

Microsoft

Other companies

👋 Mike & Erin