Category: From the Founder

Quick answer: AI fragmentation is the shift away from frontier AI being globally available to anyone who pays for it. The US government’s June 2026 directive ordering Anthropic to suspend Claude Fable 5 and Mythos 5 for all foreign nationals was the first practical demonstration of it. This post is a plain-English guide to what AI fragmentation actually means, why the market is already pricing it in, and how Australian businesses should be thinking about their AI architecture going forward.

This is not a news post. The news has been covered everywhere. This is a post about what most businesses are not seeing yet.

For the last decade, AI has felt like the internet did in the 2000s. Borderless. Available to anyone who could pay. The same tools sitting on your desk in Sydney that sit on a desk in San Francisco. That has been the deal. It has felt permanent.

That assumption has now been quietly retired.

Last Friday the US government issued an emergency directive ordering Anthropic to immediately shut off access to its newest AI models, Claude Fable 5 and Mythos 5, for every foreign national on the planet. Anthropic could not enforce that surgically, so they pulled the models for everyone. Australians, Europeans, Canadians. Switched off in hours.

If you want the full chronology, our piece on Claude Fable 5 walks through what was launched and what happened next. This post is about something bigger. Not the ban itself, but the structural shift it signals. And what Australian business owners and decision-makers actually need to understand about the world that is now forming.

I want to write this as an education piece, not a news piece. So let’s slow down and walk through it properly.

The thirty-year assumption that just broke

Here is what I think most businesses are missing.

The Fable 5 ban was not really about Fable 5. It was about a category shift in how the world treats frontier AI.

For thirty years, software has worked on one rule: capability flows to whoever can pay for it. Geography does not matter. A startup in Brisbane gets the same tools as a Fortune 500 in New York. That rule has been so consistent for so long that we have built entire business models on top of it, often without noticing.

Look at what we have done. We have moved our customer support to global SaaS platforms. We have built operational workflows on top of US AI APIs. We have made productivity assumptions that quietly depend on a single overseas provider remaining available to us. None of those decisions felt risky at the time because the rule felt permanent.

Last Friday demonstrated that the rule has changed. Frontier AI is being reclassified from a software product to a strategic asset. The same way nuclear technology is treated. The same way advanced semiconductors are treated. The same way encryption was treated in the 1990s. Governments have always controlled strategic assets. AI just joined the list.

And here is the part Australian businesses need to sit with: this is not a one-off. It is the first time a US export control directive has been used to pull a publicly deployed AI model offline. Crossed precedents do not tend to uncross.

Why SaaS is sliding and bespoke is rising

Here is something I noticed in the days following the ban that I think most people missed.

Pull up the share price of any major enterprise SaaS company over the past week. Workday. Salesforce. Oracle. They all started sliding the same week Fable 5 launched, and the slide continued through the ban. Not by huge amounts, but enough to notice. And the curves look almost identical across all of them.

That is not a coincidence. And it is not just about Fable 5.

What I think investors are starting to price in is a much bigger shift. Bespoke is becoming cheaper to build than off-the-shelf. With a model like Fable 5, a business with one technically literate person can build a custom internal tool that does what they used to pay a SaaS subscription for. The economics that made SaaS dominant for the last 15 years, which was essentially ‘it is too expensive to build your own, so just rent it’, are starting to weaken.

Mix that with the Fable ban, and you get a doubly uncomfortable position for businesses that have built operations on heavy SaaS dependency. The tools are getting easier to replace, and the tools you depend on can also be switched off. Neither of those is a five-alarm fire on its own. Together, they describe a market that is moving faster than most companies’ tech strategies are.

The Australian detail that did not make the global coverage

This is the part that hit me hardest when I dug into the timeline.

In April this year, the Australian Government signed a formal Memorandum of Understanding with Anthropic. The Minister for Industry, Tim Ayres, signed it. Dario Amodei signed it personally for Anthropic. It covered local AI investment, a planned Sydney office, research partnerships, and skills development. The Australian Government’s own announcement made a point of saying that Australians are among the highest per capita users of Claude in the world.

Two months later, on 12 June, that MoU did not matter at all.

When the US export control directive came through, every Australian Claude user was classified as a foreign national. The MoU we had signed at ministerial level did not provide protection. Being a Five Eyes intelligence ally did not provide protection. The fact that Anthropic was planning to invest commercially in Sydney did not provide protection.

I am not saying that to be alarmist. Anthropic is not the problem here. Dario Amodei is not the problem. The directive operated above the level of commercial agreements. It was a national security action that sat in a layer above any commercial or diplomatic relationship Anthropic could have built with Australia.

That is the structural insight. And it is one most Australian businesses have not yet absorbed.

What this is not

Before going further, let me be clear about what I am not saying.

I am not saying stop using Claude. We use it every day. It is a genuinely transformative tool for the businesses we work with. I am not saying Anthropic acted badly. From what I have read, they pushed back on the directive and complied because they had to. I am not saying the US government acted unreasonably either. National security decisions involve information none of us can see from where we are sitting.

What I am saying is that the assumption a lot of Australian businesses have been operating under, that access to any global AI tool is essentially guaranteed as long as we pay for it, was just stress-tested. And it failed.

The question is not whether this kind of disruption can happen. It already did. The question is what to do about it.

What Australian businesses should actually take from this

I wrote about how the businesses that adapted to the Industrial Revolution were the ones we still know today. Ford did not just buy machines. Ford rebuilt the production process around what machines made possible. The competitors who treated machines as an add-on to existing factories got crushed.

The same dynamic is happening with AI right now. The businesses that treat AI as a tool to bolt onto existing operations are going to be outcompeted by businesses that redesign their operations around what AI now makes possible. The Fable ban does not change that thesis. It sharpens it. Because the businesses that adapt thoughtfully also need to be the ones whose adaptation does not break the first time access changes.

That is the real lesson. AI fragmentation is the new structural shift. The businesses that adapt their architecture to it now will look very different in five years from the ones that do not. Building AI-resilient architecture is the Ford assembly line decision of this moment.

So what does adaptation actually look like in practice. There are three concepts worth understanding properly, because they will come up again and again over the next few years.

Concept one: single points of failure

Every business has them. Most businesses do not map them. The exercise is simple. Walk through your operations. For every process that now involves AI, ask one question: if this specific model disappeared tomorrow, what breaks. Some answers will be small. Drafting emails is fine, you can switch tools easily. Some answers will be bigger. Customer service flows, automated decision systems, content pipelines, document processing. Knowing where your single points of failure are is half the work. Most businesses skip this step because they have never had to think about it.

Concept two: model-agnostic architecture

This is the most important concept in the post. Model-agnostic means designing your AI systems so they do not depend on any single model or provider. The businesses least disrupted last Friday were the ones who had built AI automation in a way that did not depend on any single model. They could swap between Claude, GPT-4, Gemini, or open-source alternatives without rebuilding from scratch. That capability used to be a nice-to-have. Last week it became the difference between operations running normally and operations not running at all. From a design perspective, model-agnostic architecture is what protects you against three different categories of risk at once: provider outages, sudden pricing changes, and geopolitical access restrictions like this one.

Concept three: AI access as supply chain risk

After 2022, every serious business added semiconductor supply chain risk to their thinking. We learned how exposed global operations were when chip access changed. AI access risk is now in the same category, but most businesses have not added it to their risk register yet. This is the gap. AI dependency belongs in board-level risk discussions, business continuity plans, and vendor evaluations. The exact same way you would assess any other supply chain dependency in your business. The thinking model already exists. It just needs to be extended.

Zooming out: where this is heading

If you take one thing from this post, take this.

Software was supposed to be the great global equaliser. A founder in Lagos and a founder in London were meant to have access to the same tools. For thirty years, that mostly held. AI is the first technology in that lineage where the equaliser premise is breaking openly, and it is breaking at exactly the moment AI is becoming the most economically valuable thing in software.

So we get to a strange dual reality. AI is the most important productive technology of our generation. It is also the most likely technology to be controlled by governments going forward. Both of those things are true at the same time, and both will shape every business decision around technology for the next decade.

What that means in practical terms is that being a smart user of AI is no longer enough. You also need to be a smart architect of where you put your AI dependencies. Geography is back in software. Most businesses have not updated their thinking yet. The ones that do will have a real competitive advantage when the next directive arrives. Because there will be a next one.

One last thought

The Fable 5 ban itself might be reversed. Anthropic has said they believe it is a misunderstanding and they are working to resolve it. I hope they do.

But hoping that this specific incident gets resolved is not the same as understanding the world that produced it. That world is the one to plan for.

The lesson is not ‘Fable 5 got switched off.’ The lesson is that we are entering a phase where AI capability, market structure, and geopolitics are starting to move together rather than separately. Businesses that understand that early will design themselves differently from businesses that do not.

That is the entire point of this post. Not to make anyone alarmed about Fable 5. To make sure Australian business owners are not still operating on the assumption that AI access is a fixed input. It is not. And the businesses that internalise that, and design around it, will be the ones still standing when the next change arrives.

---

Fabrice Jiew is CEO and co-founder of Automation Consulting, one of Australia’s original automation consultancies. He works with businesses across retail, fintech, and professional services to build AI and automation systems that are designed to last beyond any single tool.

Building or rethinking your AI architecture? Let’s talk.

---

Next → Everyone’s talking about Claude Fable 5. Here’s what we think actually matters for your business

Common questions

Let’s be real. We ignore it as much as possible, but no matter what we call it — AI slop, people getting called out for using AI, universities banning it — AI keeps getting better. And it is happening faster than most of us expected.

If you run a business, or even just work a normal 9–5, it is hard not to feel a bit uneasy about where this is all going. Are our competitors better at using AI than us? Are we slowly falling behind? Is my job even safe long term?

If you zoom out, we have seen this before. Look at the Industrial Revolution. When machinery came in, factories that stuck to manual processes simply could not compete anymore. They were slower, more expensive, and less consistent. The ones that adopted machines just won.

Over time, “handmade” did not disappear, but it became niche. The bulk of the market — and the money — moved to industrial-scale production. But the part people do not really talk about is that the transition was not clean. People lost jobs. Entire skill sets became less valuable almost overnight. And it took a long time for new industries to catch up and absorb that workforce again. The businesses that adapted, though, are the ones we still know today.

Take Ford Motor Company. Before mass production, cars were slow and expensive to make. When Henry Ford introduced the assembly line, it completely changed the game. Cars became more accessible, demand exploded, and Ford helped define the industry.

The same thing happened with jobs. Yes, people lost them. But over time, they did not simply disappear — they shifted. A huge portion of the workforce used to be in agriculture. Now it is tiny by comparison, yet we have more total jobs than ever before. They just look completely different.

The catch is that this shift does not happen evenly. Some people adapt quickly, others do not. And that gap is where much of today’s anxiety is coming from.

At the same time, expectations changed. When things became easier to produce, people did not just buy the same amount for less. They expected more. Better. Faster. Take Coca-Cola. It did not become massive just because drinks were easier to produce. It scaled because distribution improved, branding got stronger, and demand went global.

So in reality, the jobs did not disappear. They evolved. Factory workers became machine operators, technicians, and engineers. Outside the factory, there was growth in marketing, logistics, and R&D — things that barely existed at scale before. The work became more specialised, more technical, and more important.

So we asked ourselves the same question at Automation Consulting: If we genuinely believe this shift is happening, why are we still working the old way?

Because the reality is, most companies will not change yet. They will experiment a little. Try a few tools. Maybe run a small internal project. But largely, they will keep operating the same way. And honestly, that did not sit right with us.

So we made a decision. Not to slowly phase it in. Not to treat it as an add-on. But to actually commit to it.

We have fundamentally changed how we build. We are not adding AI into our process. We rebuilt our process around AI.

At Automation Consulting, our developers are not just writing code anymore. A big part of their role now is orchestrating systems — pipelines of AI agents that generate, test, and iterate on work. Before, development was pretty linear: write something, test it, fix it, repeat. Now it is different. You design the system, let it run, validate what comes out, and refine it. It is faster, but it also changes what the job actually is.

And it is not just developers. Project managers, designers, and testers — everyone across the team is using AI in some form now. It is simply part of how we operate.

What that means in practice is that we can move faster, explore more options, and get to better outcomes quicker. We are probably one of the earlier consultancies in Australia to really commit to working this way. But globally, this direction is already happening. Companies like Stripe are clearly moving toward more AI-assisted development, where engineers can build and iterate far faster than before.

This does not mean people are out of the picture. It simply means the way we work is changing. And with that, the way businesses approach technology is changing too.

Instead of buying software and forcing your business to fit it, more companies are starting to build things around themselves — systems that actually reflect how they operate. That becomes your IP. Your workflow. Your advantage.

And I think that is really where this is heading.