2nd August, 2026
Nuggets, take two 🥚🥚
By popular demand, another slew of nuggets - politics, AI, business and sports.
Let the sunshine in
Sometimes the best new ideas are actually very old ideas.
Here’s Fred Dagg’s short sketch about the benefits of solar power, from (checks notes) 1975:1
Meanwhile, fifty years later, in the lead up to the election, both major political parties have recently announced policies that will support homeowners to install solar panels and batteries.
National: Home Energy Fund offering low-interest, long-term loans, repaid through rates (see: RNZ)
Labour: Ratepayer Assistance Scheme offering long-term, low-interest loans, repaid through rates (see: RNZ)
See the difference? 🧐
Congratulations to Mike Casey and his small team at Rewiring Aotearoa who have done an amazing job lobbying for this.
Modern monarchy
That’s all well and good. But, depressingly, it doesn’t seem to matter that much any more what the major political parties agree on. The real driver of change throughout my adult life is the now-81-year old kingmaker, first elected to parliament in (checks notes) 1978.
The first election I was old enough to vote in was the first MMP election in 1996. That was also the first time that NZ First had the balance of power. They’ve never had more than the 17 seats they won back then, but they have retained an outsized influence over the years.
I wonder how many more elections it will take for that to change?
In this environment, it should be no surprise that alternatives like the Opportunity Party are attracting so much media coverage. (So much in fact that they seem to be living rent-free in the Prime Minister’s head at the moment).
But I reckon another much bigger story is hiding in plain sight. Why is nobody talking about The Unsure Party? In recent polling they got 12% of the vote - enough, if my calculations are correct, to get 14 MPs. I bet most of you couldn’t even name their leader!
It reminds me a little of the student council elections when I was at university, where you could vote either for or against a candidate. Perhaps that’s the electoral reform we need to unclog things?
Usain Boat
Speaking of Winston: One of the projects he’s been responsible for in this soon-to-be-completed parliamentary term is the replacements for the replacement Interislander ferries. He recently held a press conference to announce these two boats would be called Cook and Kupe.
I know … we’ve been here before but … I think we could do so much better.
Here are twenty-odd alternatives from a quick brainstorming session, starting with a few that might appeal specifically to old-mate:
Benson & Hedges
Whisky & Soda
Laphroaig & Talisker
Devine & Bates
Back & 4th
Iwi & Kiwi
Jules & Linda
Dairy & Tourism
Bilbo & Frodo (in for a penny, in for a pound, and all)
Bret & Jemaine (leaving the door open for Murray if we can ever afford a third)
Wellington & Picton (doubling down on the Cook reference)
Hudson & Halls (to appeal to the boomers)
B1 & B2 (to appeal to the millennials)
All Boat & Boat Fern
Boult & Southee
Insult & Injury
Neil & Tim
Trevor & Shirley
North & South (after the islands, not the magazine)
White & Brown (for diversity)
Marmite & Vegemite
Up & Under (might be too close to the bone, given reliability record of the current ferries these two would replace)
Deep & Wide
Red Peak & Laser Kiwi
The Cake Tin in Wellington might officially be Hnry Stadium these days, but it will always be The Cake Tin. We can call these boats whatever we want.
(see: Boaty McBoatface, for those who don’t get the Usain Boat reference)
Just send me the prompt
This is the modern way: Take three bullet points of actual thought, ask an AI to inflate them into five paragraphs of professional-sounding padding, and then the recipient asks another AI to compress them back down to three bullet points.
“If you’re going to use an LLM to write me an email, I’d much rather you just send me the prompt; at least then I’d have an idea of what you actually meant to say.”
What a time to be alive!
Patronage
There are a number of different job titles that we give to people who build software. I’ve had most of these: programmer, developer, engineer. And let us never forget: webmaster.
The fashionable labels now, given the growing prominence of tools like Claude Code and Codex, are orchestrator or conductor or (my preference to describe how I now build software): director. These at least admit that we humans are no longer playing the instruments or speaking the lines. We are all code reviewers, testers and product managers (at best) now.
Here Ethan Mollick suggests the next rung on that ladder, after spending time with the latest generation of models:
“Last year I called this working with a wizard: you chant the spell and something happens. With [the latest models] the spell has gotten powerful enough that I am no longer sure I am the wizard. I am closer to a patron. I describe what I want, I pay for it, and I judge the result.”
A patron commissioning work actually feels about right to me. Whether that’s comforting or terrifying probably depends on how good your taste is.
The Medici never picked up a paint brush, but we still remember whose name was on the invoice.
Weight bearing words
Which begs the question: If you use an AI to generate content, who is responsible for it?
The answer seems simple to me. If you put your name on a piece of work, you are claiming responsibility. The tools you used to create it are not really the point, unless those expose you as not really having claim to the work in the first place.
Some examples:
If you’re a developer (or director!) and deploy software that you can’t understand or haven’t bothered to test, then you are responsible for whatever that code does.
If you’re a lawyer and you make a submission to the court that contains hallucinated case references, you are responsible for the impact of that on your client.
If you’re a student and you submit an AI generated essay, and as a result you don’t really understand the content that essay was intended to test, then don’t be surprised if you fail the course.
(Although, in that last case, I’d argue if you’re a teacher who still believes that asking for an essay is a good way to assess understanding then you should probably also ask for those essays to be submitted in cursive.)
Matt Webb, in a post about surrendering his inbox to an AI agent while under siege from FedEx customs forms, having apparently been mistaken for a watch smuggler, has the best description I've read of why we should be wary of letting the machines do all our writing:
“I got scared off using ChatGPT to help with my blog pretty early when I was talking through an editing decision and it came up with a turn of phrase that was so perfect and so unique that I couldn’t resist it. But it didn’t represent any thinking that I had done to arrive at it, this perfect metaphor, so it wouldn’t bear my weight when I leant on it.”
Writing is thinking.
If you didn’t do the thinking, the words are scaffolding around an empty building.
Omit needless words
Of course there are many tools that can be employed to improve writing, after you’ve done the thinking. For example, Strunk & White's most famous rule is now available as a robot that asks an AI which words in your text can be deleted without changing the meaning.
Rare candy
Last time I published a list of Nuggets, I recommended Nik Wakelin’s Startup Theatre podcast interview.
Here he is again, this time speaking at a recent Auckland AI meet-up on power-levelling spreadsheet agents (you’ll have to click the link as the video has been disabled for playback on third-party websites). Come for the Pokémon references. Stay for the hard-earned lessons about what it’s actually like to “direct”.
Tomasz Tunguz argues that software engineering is no longer about building the interface or managing the data; it’s about building the harness - the constraint-plus-reliability layer that turns an untamed model into a dependable workhorse.
Most of what I know about agentic engineering I’ve learned by watching Nik and Ludwig and the rest of the Sterling team working at a mind-blowing pace.
Stay tuned…
Asking for a friend
Tim Ferriss, armed with his own data, asks: Has AI already killed how-to nonfiction?
His sales: down 5% in 2023, 13% in 2024, 46% in 2025, and trending even worse this year.
“In 2019, the best interface to those answers was a book. In 2026, millions believe that the best interface is a free chatbot that has read my books — and thousands of others — that will give you a personalized protocol in 15 seconds.”
My book was published in February 2025, so this headline caught my attention.
This is my working theory (hope?): an LLM can summarise a protocol, to use Tim’s expression, but it can’t spend years in the rooms where the stories happened.
Information is now free-er. Scar tissue is still full price.
Castles in the sky
Tomas Pueyo doesn’t invest in real estate, and his reasoning should cause discomfort for many: ever-rising house prices are a post-1950 anomaly, not a law of nature. Before then, real prices were flat for centuries.
“We’re moving from a world where people have only experienced growing housing prices, to one where they are likely to shrink.”
Every demand driver is stalling; supply constraints can only loosen.
This is a dangerous idea to hold, let alone say out loud, in a country that treats housing as its national retirement scheme.
Related: this reel made me laugh: “they could just… live in it”.
Goodbye success fees
Trade Me has scrapped success fees for casual sellers:
I’ve seen this described as “the biggest change to the marketplace in its history”. I’d argue it’s the second-biggest change, after the change we made to introduce fees for the first time, back in 2001.
This is obviously a response to Facebook Marketplace. Possibly 10 years too late?
Note the mechanics though: sellers now keep 100%, buyers pay a service fee, and payments carry their own clip. Fees, like energy, are neither created nor destroyed; they are merely relocated.
Related: A question for you all to ponder: Is Trade Me a “tech” company? Sam Morgan was added to the (suddenly-contentious) Hi-Tech Hall of Fame way back in 2016, but I think I’m correct to say the company has never been counted in the TIN Report. ¯\_(ツ)_/¯
[Unpaid Ad]: Smails
Smails: every Gmail and Outlook account in one fast native window, no browser tabs, no subscription - $9.99 once, yours for good.
Developed by Matt Allen - co-founder at Tractor Ventures, where we are early investors - mostly to scratch his own itch.
It’s interesting partly because it’s a useful tool, partly because it was built so quickly using AI, and partly because “pay once, own it” is these days a radical business model.
Football was the winner?
I reluctantly admit: I actually really enjoyed the recently completed FIFA World Cup.
I say “reluctantly” because I started out deeply cynical about the expanded 48-team format. It was great for the All Whites, who (for the first time) had a direct qualifying route as the best team from the Oceania Confederation. But 12 groups, 104 matches and notably five knock-out rounds, one more than in any previous tournament … I wasn’t sure that would suit the best teams, since they had many more opportunities to trip up against minnows or less fancied teams.
What actually happened is: the best teams won.
That might sound strange to say, but one of the things that is remarkable about football is the best team doesn’t always win. In fact, often the best team doesn’t win.
There is a moneyball metric we can use to dig into this: xG measures the quality of a chance at the moment the shot is taken. Each shot gets a probability of scoring (0 to 1) based on historical data from thousands of similar shots: distance and angle to goal, body part used, type of assist (through ball, cross, or cut-back), defensive pressure, whether it's a one-on-one, rebound, etc etc.
So a penalty is ~0.76 xG; a hopeful 30-metre punt might be 0.02. When we take the sum of all of the shots taken during a match we get the team's xG. This is effectively "how many goals would an average finisher score from these chances", discounting finishing luck and heroic goalkeeping.
Here’s an analysis of all 31 knock-out games.
Round of 32
Canada 1–0 South Africa — xG 1.38–0.14 ✅
Brazil 2–1 Japan — xG 1.72–0.23 ✅
Paraguay 1–1 Germany (pens 4–3) — xG 0.42–1.49 ❌ Germany robbed
Morocco 1–1 Netherlands (pens 3–2) — xG 1.38–0.24 ✅
Norway 2–1 Ivory Coast — xG 2.02–1.36 ✅
France 3–0 Sweden — xG 3.17–0.66 ✅
Mexico 2–0 Ecuador — xG 1.05–0.75 ✅
England 2–1 DR Congo — xG 2.04–0.80 ✅
Belgium 3–2 Senegal (extra time) — xG 1.74–3.58 ❌ Senegal robbed
USA 2–0 Bosnia & Herzegovina — xG 0.88–0.25 ✅
Spain 3–0 Austria — xG 2.80–0.29 ✅
Portugal 2–1 Croatia — xG 2.18–1.34 ✅
Switzerland 2–0 Algeria — xG 2.45–0.74 ✅
Egypt 1–1 Australia (pens 4–2) — xG 1.32–0.84 ✅
Argentina 3–2 Cape Verde — xG 2.16–0.45 ✅
Colombia 1–0 Ghana — xG 2.04–0.27 ✅
Round of 16
Morocco 3–0 Canada — xG 0.82–0.84 ➖ dead even, but not on the scoreboard
France 1–0 Paraguay — xG 1.45–0.13 ✅
Norway 2–1 Brazil — xG 1.05–2.61 ❌ Brazil robbed
England 3–2 Mexico — xG 1.55–1.94 ❌ England nicked it?
Spain 1–0 Portugal — xG 1.77–0.60 ✅
Belgium 4–1 USA — xG 2.15–0.67 ✅
Argentina 3–2 Egypt — xG 2.80–0.98 ✅
Switzerland 0–0 Colombia (pens 4–3) — xG 0.35–1.03 ❌ Colombia robbed
Quarterfinals
France 2–0 Morocco — xG 3.69–0.14 ✅
Spain 2–1 Belgium — xG 1.96–0.34 ✅
England 2–1 Norway — xG 1.04–0.68 ✅
Argentina 3–1 Switzerland (extra time) — xG 2.00–0.53 ✅
Semifinals
Spain 2–0 France — xG 1.63–0.31 ✅
Argentina 2–1 England — xG 1.59–0.53 ✅
Final
Spain 1–0 Argentina (extra time) — xG 1.94–0.20 ✅
There were arguably only four upsets in all of those matches, and nothing beyond the quarter-finals.
Based on the pre-tournament rankings, the semi-finals were 1 v 4 and 2 v 3. The final was 1 v 2. And the winner was Spain, the top-ranked team.
That’s how a tournament is supposed to work!
The runs aren’t yours
On Kane Williamson’s recent mid-series retirement, 485 test runs short of 10,000, Tim Grgec is gently devastating:
“The retirement, of course, was his to make. But the runs belonged to us, too.”
(writing on Dylan Cleaver’s excellent The Bounce)
Dylan himself has described Williamson as the Platonic ideal of a New Zealand icon: understated, humble, team-first, deflecting credit as adroitly as he deflected good-length balls to the third-man boundary. Choosing serenity as his superpower. And yet, with a busy summer of test matches ahead, including a tour to Australia and a Boxing Day test at the MCG, retirement feels less like a fitting ending than the culmination of his stubborn modesty. Final proof of how much he cares about not caring. The canonical #QuietOne!
Stepping back, though: what a run it has been to be a New Zealand sports fan. The Richie McCaw era flowing more-or-less directly into the Kane Williamson era, with Lydia Ko and Ryan Fox making it seem normal to win at genuinely global sports, and Lisa Carrington and others compiling outrageous careers of their own, winning more Olympic medals than former New Zealand teams ever dreamed possible … all in parallel.
We probably won’t appreciate it properly until it’s over. Is it over?
Duck
In 195 test innings across 110 matches Kane scored a duck (zero runs) 12 times.
Even Don Bradman made a duck seven times in his 80 test innings. Nearly 9% of the time, the greatest batter who will ever live failed to trouble the scorers - including, famously, his final innings in 1948: bowled second ball, needing just four runs for a clean career average of 100.2
Perhaps the rest of us can relax about our failure rate?
Also, just 80 innings, across a twenty-year career - Neil Wagner and Ian Smith both had more. Comparing raw totals across eras is a nonsense.
Wet words
Matt Webb, again, in a post titled Wet thoughts, with my new favourite retronym: wet signature - i.e. one applied with actual ink, by an actual hand.3 Apparently there are still situations where this matters in terms of authentication.
“But from now on, I guess, most words will be words not [written] by humans, and that’s the new default. So we’ll need a name to specifically mean human words. Wet words? These are wet words!”
Meanwhile, one of my guilty pleasures is seeing the lazy journalism tropes roll out every single time the All Blacks play: the player ratings (a number from 1-10, assigned to eighty minutes of chaos), normally closely followed by the “world media reacts” roundup (often not much more than three tweets and a Daily Mail headline).
Content slop predates AI. We write it ourselves, with our own hands.
Wet slop, if you will.
Now and then
“Few people think more than two or three times a year. I have made an international reputation for myself by thinking once or twice a week.”
— George Bernard Shaw
No notes.
My book, How To Be Wrong, is now available on Shop.app To celebrate save 10% this month if you use the discount code TOPTHREE. Enjoy!
Photo by Zunaira Bilal Anjum on Unsplash
Bradman’s final average of 99.94 has a sibling. Denise Annetts averaged 81.90 in women’s test cricket - the only other average, in either game, even in the same ballpark (or oval?) See: Women’s Test cricket
This bumps my previous favourite retronym, the muscle bike (i.e. a bicycle powered by a human battery), into second place. Like “landline”, a retronym only shows up once the old default has quietly stopped being the default. What else will shortly be on this list?