New NZ AI Public Research Org Opens

On Friday I heard the usual lines about AI: "it makes things up", "the citations were wrong", "the numbers come back different every time". They're right, and that is what happens.

Then it clicked in me. Most people judge AI by the model, and the model on its own guesses, forgets, and makes things up. The people who rely on it and get immense value from AI are using the system (sometimes called a harness) around it: tools it can run, reliable context files, a memory that survives, and code it writes and then runs. They use the same models, in a totally different machine.

Almost every frustration people have with AI has an engineering fix, and it's more achievable to overcome than most think. I want to close that gap in the open, on real problems, not polished demos.

So, a new segment: send me a task you're working on with AI and the problem you're hitting, with a specific or an example so I can diagnose it. It could be as basic as:

I'll unpack the problem, build a solution, and share a short video of how it works and how it was built (happy to keep it anonymous if you'd prefer).

Reply with your challenge, and I’ll assess it and get building.

Happy reading ✌️

Did someone forward you this? Sign up!

Our take: NZIAT lands just as AI moves from hype into real business use, so the opening for research-to-firm partnerships is wider now than it would have been a year ago. The open question is where the AI slice of the $71m actually goes, because a right-to-win strategy means backing a few industries rather than spreading thin. If access flows to the players with existing relationships and resource, the local playing field doesn’t really even.

Our take: Mercury's 12.7% stake ties a listed power company directly to AI demand, which is a new kind of exposure for the NZ grid. If the AI compute bubble deflates, the loss does not stay offshore, it lands on a domestic generator's books.

Our take: Data-centre governance is being written council by council rather than nationally, which means an operator faces a different rulebook in Waikato than in Southland. That patchwork is exactly what a national framework was meant to prevent, and it is forming anyway. Being named an "advantaged zone" by Invest NZ is what pushed the council to act, so central government's promotion is creating the local caution it now has to manage. The campaign and the pushback are pulling against each other inside the same government.

Our take: The objection will be that this pushes tools like ChatGPT onto students, but students are already using them, and short of banning the phones and laptops they run on, a school cannot switch that off. Built-in controls at least put some guardrails around use that is happening anyway. Overseas, AI is being taught inside the curriculum while New Zealand is still debating whether to allow it, so a Kiwi student left to figure out ChatGPT alone arrives in London or Sydney noticeably behind. The skills are in hot demand in the markets our young people move to.

Our take: Students save time with AI, but the work reappears in teachers' marking queues. Any national framework needs to account for that transfer of workload. Teacher Sonya McAteer says very few students intend to cheat. Dependence is harder to address because detection and punishment do little to rebuild independent thinking.

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

1️⃣ What Claude's new text watermark actually means for businesses: Two board papers carry the same Claude mark: one built on a week of interviews, the other a first response nobody checked. Provenance is not performance. Businesses that treat the mark as a verdict reward hiding AI, not using it well.
Read.

2️⃣ What can NZ businesses learn from AWS's AI rollout? Not the 4,000 MCP servers or 40,000 field staff. The transferable moves sit underneath: pick one customer outcome, settle who owns its data, redesign the work before adding AI, then measure the outcome, not tool usage.
Read.

3️⃣ Deloitte found 16% of leaders think their processes are ready for agents: Dig into any workflow and the answers stop: people describe the same step differently, one step lives in someone's head, the data sits in systems that do not connect. Dropping an agent on that produces faster confusion, not value.
Read.

4️⃣ The common advice on teaching AI about a business is flawed: Everything fed to AI sits in its context window, and too much drags the output down. One job per file, a header saying what is inside and who owns it, the unwritten rules written down, dead versions deleted, because two files that disagree beats one slightly wrong.
Read.

5️⃣ Point AI at where the thinking happens, not where the data ends up: Why a customer said no, why a team changed direction, and the workaround nobody documents. Most of it gets said in meetings then disappears. Capturing context isn’t sufficient enough to create a competitive advantage moving forward.
Read.

The selected top headlines from each major AI tech company.

Anthropic

OpenAI

Google

Chinese Frontier Labs

Open-Weight Model Updates

xAI

Microsoft

It’s Mike here, I run The AI Corner.

I'm not just into writing about AI. I run Allexive. We work with businesses that have spent on AI licences and training and still can't point at what it changed. We redesign the work so it does.

Let’s chat if you’re interested to learn more →

👋 Mike & Erin