Anthropic board director: NZ has 5 years, not 20

📢 OFFER: 1:1 Done-With-You Claude Code Coaching for non-technical knowledge workers with Mike.

You're drowning in information. Too many meetings, too many emails, too much to read, too much to decide. AI was meant to fix this.

Custom GPTs, saved prompts, Claude Project folders... You've done more with chat-based AI than most, and you've started to feel like you've hit the ceiling.

But here's what ChatGPT and Claude still can't do for you today:

❌ Remember who you are from one chat to the next
❌ Seamlessly move outputs across your systems without copy-pasting
❌ Run your workflows autonomously
❌ Fully wire into the rest of your tech stack
❌ Actually read, write, and organise your own files
❌ Behave like anything more than a clever chat window

Claude Code is what's next. Not just for building apps, but for managing knowledge work as a non-technical operator.

I'm opening three seats in a private 1:1 pilot to take knowledge workers through the four-week journey that took me six months to figure out.

Together we will build an AI system that actually compounds its learnings: the exact same setup I use to power my own sales, marketing, and operations.

Video of the week: You know it’s a worry when this type of content resonates with your wife! Time to start setting stricter personal boundaries…

Happy reading ✌️

Did someone forward you this? Sign up!

Anthropic board director and NEXT Foundation chairman Chris Liddell warned in an NZ Herald opinion piece that New Zealand, with "our often-characteristic inertia", risks being overwhelmed by the speed of AI progress. Liddell proposed an AI Talent Visa, a Sovereign AI Infrastructure Fund using NZ's renewable energy, and a call for national urgency. The following day, Simplicity KiwiSaver boss Sam Stubbs pushed back, arguing NZ does not have enough clean energy for its own needs, let alone to power overseas AI compute. 3 min read.

NZ government agencies have spent at least $20.2 million on AI and automation tools since January 2023, but claimed savings are self-reported and sometimes calculated using tools provided by the vendors selling the products. Part 3 of BusinessDesk's investigative series by Cecile Meier found ACC has four AI tools embedded across its claims pipeline, though staff always make the final call. Researcher Kyle Higham warned AI tools could lead to staffing cuts and worse public service delivery. 4 min read.

NZ deep-tech startup Scentian Bio secured $7M in an oversubscribed pre-Series A led by Icehouse Ventures, bringing its AI-powered biosensor that uses insect olfactory receptor proteins to detect volatile organic compounds with femtomolar-level sensitivity closer to market. The Gates Foundation has funded Scentian Bio for non-invasive breath-based detection of tuberculosis, cancer, and COVID-19. Near-term commercial focus is food quality, with a Zespri pilot underway. 3 min read.

A company conducted more than 1,000 AI-powered interviews with New Zealanders in a single month, according to Stuff, as the rapid scaling of AI recruitment tools draws legal warnings from Simpson Grierson. The law firm cautioned that NZ employers remain legally liable for AI-driven recruitment errors, bias, and misleading information. If AI screening disproportionately filters out candidates by protected characteristics, it amounts to unlawful discrimination. Spark NZ uses an AI "smart interviewer" for contact-centre hiring. 3 min read.

Mike’s twisted Justin’s arm and secured AI Corner readers a $50 discount off Roadshow tickets. Use the discount code AI-CORNER when you buy a pass to any Roadshow event.

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

1️⃣ The AI conundrum: 1,000 x $10k wins sit untouched while leaders hunt 10 x $1m plays: Most businesses collapse three separate questions (what to apply AI to, how to build it, where value lands) into one committee conversation, and nothing ships. Separating productivity AI from engineered AI, and chasing the thousand small wins first, is what breaks the stall.
6-min read.

2️⃣ Why benchmarking GPT vs Claude vs Gemini for 3 months is a bikeshed: A retail team spent a quarter comparing models for product recommendations instead of testing the workflow itself. At proof-of-concept stage, model choice barely matters; the real questions (what problem, what good looks like, who can tell the difference) are the ones teams retreat from into comparison spreadsheets.
2-min read.

3️⃣ How my content operating system hits 38/40 on a programmatic linter before a human reads it: A Python script checks 136 banned phrases across 19 categories, scores eight passes per draft, and auto-fixes violations before anything lands for review. The build is modular markdown and JSON in an Obsidian vault, written by Claude Code, and the hard part was never technical ability; it was codifying taste into rules a machine can enforce.
6-min read.

4️⃣ Claude Code fills in 15-field forms in under a minute by reading answers from Google Drive: The real time cost of forms is not filling one in, it is the repetitive loop of retrieving the same personal and company details across insurance claims, supplier onboarding, and tax paperwork. A context file plus a hook that auto-saves new details turns each form into training data for the next one.
2-min read.

This week's guest is Dan Browne, owner of Indelible with 16 years in video production. Hear:

🎧 Listen on Spotify or YouTube.

Helping advanced builders and tinkerers stay at the frontier of AI.

What I'm researching

1️⃣ Anthropic's Managed Agents is the one I can't stop thinking about this week. In plain English: until now, shipping a real agent for a business (one that reads emails, updates a CRM, runs reports, sends follow-ups) meant infrastructure work before the agent did anything useful. Managed Agents takes all of that off the table. You define the job, Anthropic runs it on their cloud, the agent gets a secure workspace with code execution, web access, and tool connectivity built in.

My thesis on why this matters for NZ: we're already seeing Anthropic tooling adopted by tier-1 companies and SMBs alike via Claude Code. Engineering teams land first, build internal agents on this stack, then pull business functions in behind them (IT will only let them build in this environment). Managed Agents is the platform layer that makes that inevitable. If you're deciding where to build enterprise agents, this is the substrate to bet on. (announcement)

What I'm workshopping

2️⃣ LangChain's piece on continual learning for agents: the argument is that the missing layer in most agent stacks isn't better models or better retrieval, it's a mechanism for the agent to actually learn from its own runs and carry that forward. Worth a read if you're trying to make your Skills or agents compound rather than plateau. (link)

Handy repo

3️⃣ A repo that turns anything into clean markdown for LLMs. Shared by @_vmlops, the point is that most of what you feed an agent (PDFs, web pages, docs, slide decks) is still trapped in formats that burn tokens and blur structure. A single CLI that standardises everything into well-formed markdown is one of those small tools that quietly fixes a lot. I've been moving my own wiki and capture pipeline toward markdown-everywhere for 6-months for exactly this reason, it's the lowest-friction substrate for agents to read, search, and recompose. (link)

The selected top headlines from each major AI tech company.

Anthropic

OpenAI

Meta

Google

Microsoft

Perplexity

It’s Mike here, I run The AI Corner.

I’m not just into writing about AI. I help businesses grow without adding headcount by shipping AI systems that cut costs and increase throughput.

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

👋 Mike & Erin