Are you found by AI? Log in Start free trial →

← Tech Horizon News

Issue 30 · · sent to our list

The Australian Government just signed its first AI deal. I built a

Week of: April 7–11, 2026 (SEQ Edition) Reading time: 5 minutes

The Australian Government just signed its first AI deal. I built a product on a Sunday afternoon that got 20,000 impressions with $50 in ad spend. And OpenAI cut its business pricing to $20/seat.

Big week. Let's get into it.


🇦🇺 Anthropic Signs Deal with Australian Government

This one matters.

Last week, Anthropic CEO Dario Amodei flew to Canberra and signed a Memorandum of Understanding with the Australian Government — the first deal under Australia's National AI Plan.

Here's what's in it:

$3 million in Claude API credits going to four Australian research institutions — ANU, Murdoch Children's Research Institute, Garvan Institute, and Curtin University.

A startup credit program offering up to USD $50,000 (roughly AUD $72,000) in API credits for VC-backed deep-tech startups. Drug discovery, climate modelling, materials science, medical diagnostics.

A Sydney office confirmed for 2026.

And something your competitors aren't talking about: Anthropic will share its Economic Index data — tracking how AI is actually being adopted across Australian sectors including healthcare, agriculture, resources, and financial services — directly with the government.

Their data shows Australians use Claude across a wider range of tasks than most other English-speaking nations. High-skill applications. Management. Sales. Business operations.

Why this matters for you: The tool I've been teaching in the Academy for two years just signed a deal with your government. That's not hype. That's infrastructure. If you've been on the fence about building on Claude, the fence just moved.


🔨 I Built PaddockMap in 3 Hours. The Government Spent $3.5M.

The Australian Government is spending $3.5 million on a food supply chain study. Results due 2027.

Last Sunday afternoon, I sat down with an AI coding tool and built PaddockMap.com.au — a local farm and food discovery platform that lets you find farms, co-ops, markets, butchers, fishmongers, and bakeries near you.

Three hours. Sunday evening. No engineering team.

Within 24 hours: 20 users on the platform.

Across social media: 20,000+ impressions. $50 on LinkedIn. Zero on everything else.

This isn't theoretical. This is a live product, with real users, solving a real problem that a government department allocated $3.5 million and 18 months to study.

Same week: I republished techhorizonlabs.com using Replit. 100% scores across all four PageSpeed Insights categories — Performance, Accessibility, Best Practices, SEO. Entirely vibe-coded.

For context: a perfect 100 puts you in the 98th percentile of all websites globally. Most agency-built sites don't hit it.

The honest caveat: High PageSpeed scores don't mean production-grade security. AI-generated code gets you fast, accessible, well-structured websites. But if you're handling customer data or payments, you still need a security review. Speed and security are different conversations.


🔒 Claude Mythos and Project Glasswing — Why Your Security Baseline Matters Now

Remember when I mentioned Mythos at our roundtable two weeks ago? Fortune broke the full story on Monday.

Anthropic gave early access to its unreleased flagship model — Claude Mythos — to AWS, Apple, Google, Microsoft, CrowdStrike, Nvidia, and JPMorgan. The program is called Project Glasswing, and its focus is defensive cybersecurity.

What they found: thousands of zero-day vulnerabilities, including bugs that had gone undetected for 27 years. Cybersecurity stocks dropped 5–11% on the news. Expected public launch: Q2 2026.

What this means for you: AI is finding vulnerabilities faster than humans can patch them. If you haven't set up your security baseline — passkeys, a password manager like Keeper, backup software, multi-factor authentication — now's the time.

Our next monthly roundtable includes a guest security auditor. If you're an Academy member, that session just became essential.


💰 Price Watch — Action Items With Deadlines

OpenAI drops ChatGPT Business to $20/seat

Permanent price cut. Was $25, now $20/seat/month (roughly AUD $31). Plus a new pay-as-you-go option for Codex (their AI coding tool) — no seat fee, just pay for what you use. There's up to $500 in promo credits for new Codex team members.

If you have a dev team: The $500 credit makes piloting Codex essentially free. Worth testing this month.

Anthropic blocks third-party tools from Claude subscriptions

If you use OpenClaw or similar agent frameworks with your Claude Pro or Max subscription — you need to act before April 17. Anthropic is ending flat-rate access for third-party tools. After that, it's API rates ($3/$15 per million tokens for Sonnet, $15/$75 for Opus).

For many users there should be extra credits to claim for free if you head to your “Usage” settings and just click “Claim”.

There's a one-time credit equal to your monthly plan cost, redeemable until April 17. Don't miss it.


Deadlines


🧰 Vibe Coding Tools — What to Actually Use

I keep getting asked "which tool should I start with?" Here's my current take, tested across real projects:

| Tool | Cost (USD/mo) | Skill level | Best for | | --- | --- | --- | --- | | Lovable | Free–$50 | Low | MVPs with clean design, limited integrations/ hosting | | Bolt.new | Free–$25 | Low | Quick prototypes, painful back end | | Replit (Get 1 month free here) | $20–100+ | Low-Med | Full lifecycle (code + host) | | v0 (Vercel) | Free–$30 | Medium | UI/frontend components | | Cursor | $20 | High | Production code | | Google AI Studio | included with Google Workspace | Low-High | Production code, with Google services integrated |

The workflow I recommend: Prototype in Lovable or Replit → validate with real users → rebuild in Cursor if it needs to scale.

Replit warning: Credit costs are unpredictable. Users report $100-300/month bills on a $25 plan. The agent can get stuck in debugging loops that drain credits. Budget accordingly or use Claude Code inside the Replit Shell.

The market validation is real. Lovable raised $330M at a $6.6B valuation. Cursor hit $2B ARR by March 2026. This is an entirely new category.

Excerpt from our AI Readiness Report found on our website here


🏆 Academy Wins

Real results from real members this week:

A local education business used Claude to process invoices for the first time. Their words: "Claude's done the most amount of work he's ever done for me." Upgraded to Max plan ($150/month) for end-of-month workload peaks.

Joanne Hill is learning to build a horse riding event scoring app using Replit. No coding background.

I'm building an airport transfer and car rental management app for a client. Again — no traditional dev team.

600+ members in the Facebook group. Just signed 3 Partner Tier contracts.

The pattern holds: the businesses winning are treating AI as an operating system, not a tool they open sometimes.


⚖️ December 2026: Your AI Compliance Deadline

This one's flying under the radar and it shouldn't be.

Australia's Privacy Act changes take effect in December 2026. New transparency obligations around automated decision-making.

If your business uses AI for decisions that substantially affect people — hiring, lending, insurance, customer analytics, service eligibility — you'll need to disclose when those decisions are automated.

There's no standalone AI Act. Australia is folding AI regulation into existing laws: Privacy Act, Consumer Law, Online Safety Act. Five regulators now have a piece of the pie: OAIC, ASIC, ACCC, APRA, and TGA.

What to do now: Start documenting which decisions in your business involve AI. Where does it recommend? Where does it decide? Where does a human review? Eight months is less than you think.


⚡ Quick Hits


🎯 Your 3-Step Action Plan

  1. Try a vibe coding tool this week. Open Lovable.dev or Replit. Describe something simple your business needs — a calculator, a booking form, a landing page. See what comes out in 10 minutes. Free tier works.
  2. Check your Claude subscription. If you use any third-party tools through it, redeem your credit before April 17. Check which model versions you're on — Haiku 3 and 1M context beta are both sunsetting this month.
  3. Start a simple AI decision log. One Google Sheet. Columns: Decision type, AI tool used, data inputs, human review (yes/no), date. You'll thank yourself in December.

💬 Want to Talk About This?

I do 5-minute chats with business owners who want to figure out where AI fits. No pitch, no prep needed. Just tell me what you're working on and I'll point you in the right direction.

Book a 5-min chat →


Not ready for a call? Start free.

1,190+ prompt templates. 214+ vetted tools. HuxleyGPT — your AI business advisor. No credit card.

Get free access → techhorizon.academy


The gap between "I should try AI" and "I shipped a product with AI" is about three hours on a Sunday.

Close the gap.

— Huxley

techhorizonlabs.com

P.S. Comment BUILD on the social post for this newsletter and I'll send you the Vibe Coding Starter Kit — tool comparison matrix + your first project checklist.

Get the next one

One email most weeks. Inside each issue:

We confirm by email first, so nobody gets added by someone else, and every issue carries an unsubscribe link.

Previous issue#29 A solo consultant can now operate like a five-person team. ThreeNext issue#31 Anthropic caught up to OpenAI this fortnight. Almost nobody's noticed
Method. Found by AI asks ChatGPT, Claude, Perplexity and Google's AI the questions real buyers ask for each category, then measures which businesses the engines actually name. Scores are 0–100 point-in-time measurements; the index re-scans and grows automatically. Businesses listed here were selected and measured by us from public directories, scan your own site free to see where you'd rank. Queries and methodology are open: github.com/techhorizonlabs.