AI Could Change Everything. Just Not as Fast as Investors Think.

I work with AI every day.

I see its power most clearly in software. The speed at which an idea can move from a prompt to working code is extraordinary. I also see the promise in creation and building more broadly — in writing, design, research, analysis and planning.

But I also see the limitations.

Outside software, AI still feels much more assistive than transformative. It can draft, summarise, suggest and analyse. It can make people faster. But in many industries it has not yet fundamentally changed how the work gets done, how companies are structured, how people are paid, or how customers buy.

That tension came up repeatedly when I was coaching the Australian Technologies Competition finalists. Across very different companies and sectors, the same theme kept returning: AI is clearly changing what is possible, but the path from technical capability to real commercial transformation is still uneven.

That leaves us in a strange place.

We seem to be living in a world of boom or doom. On one side, AI is about to transform every industry, replace huge amounts of knowledge work and create enormous economic value. On the other, jobs disappear, professions collapse and existing business models are swept away.

And between those two extremes, there is a lot of hype.

So what actually feels real?

I think the best way to answer that is to look backwards before we look forwards.

AI is not the first disruptive technology to arrive with extraordinary capability and enormous expectations. The microchip, the PC, the internet, cloud computing, the smartphone and electricity all changed the world. But none of them changed it overnight.

A gap always existed between invention and transformation.

The technology arrived first. Then people learned how to use it. Organisations adapted. Workflows were redesigned. New products and business models emerged. Only then did the deeper economic disruption become obvious.

That makes the adoption curve a useful lens for AI today.

Where are we really on that curve? Where has AI moved beyond assistance into genuine production and execution? What can software development teach us about what happens next? And what signals should we watch for in other industries to know when real transformation has begun?

That is what I want to explore.

▶Watch on YouTube

Open the companion video directly on YouTube.

AI Adoption Is Not the Same as AI Disruption

One of the biggest mistakes in the current AI conversation is treating adoption and disruption as the same thing. They are not.

A company can adopt AI without changing very much at all. A lawyer can use ChatGPT to draft a memo. A recruiter can use AI to write outreach. A marketer can generate copy. An accountant can summarise a report. All of that is useful. It may save time and improve productivity, but it is not necessarily disruption.

Real disruption begins when the technology changes the structure of the work itself. It changes the workflow, who does what, how long something takes, the economics of delivery, and eventually customer expectations.

That distinction matters because AI adoption is already widespread, while genuine structural disruption is still much more uneven.

Software development is probably the clearest example of where the shift has gone further. A few years ago, a developer might write code in an editor, search Google, read Stack Overflow, work through documentation, debug line by line, test, deploy and repeat.

Then AI copilots arrived and helped with parts of that process.

Now tools such as Cursor, Codex and Replit are beginning to change the shape of the work itself. The developer increasingly starts with a goal, provides context and constraints, and lets an AI system generate code, inspect files, run tests, fix problems and iterate.

The human is moving up a level. Instead of manipulating every part of the artifact directly, they are increasingly orchestrating the system that produces the artifact.

That gives us a useful way to think about AI adoption:

Assist → Produce → Execute → Restructure

In the assist stage, AI helps a human do the work. In the produce stage, AI creates a meaningful work product. In the execute stage, AI starts completing parts of the workflow itself. And in the restructure stage, the organisation redesigns the work around that new capability.

Most industries are still somewhere between assist and produce. Software is one of the first places where we can clearly see the move into produce and execute.

That matters because it gives us a preview of what disruption may look like elsewhere.

The question is not simply whether lawyers, accountants, recruiters or marketers are using AI. The better question is whether they are still using AI as a tool, or beginning to hand over the work itself.

That is where adoption starts to become transformation.

What Software Is Teaching Us

Software development is useful because it shows us what AI disruption looks like when it moves beyond assistance.

The old model was very tool-centric. Developers wrote code in editors, searched Google, read Stack Overflow, checked documentation, debugged problems, tested changes and deployed them. The human operated directly on the artifact.

Then came copilots. AI started suggesting code, completing functions and helping with debugging. That was a meaningful productivity gain, but the basic workflow still looked familiar. The developer was still doing the work, just with better assistance.

Now the model is shifting again.

With tools such as Cursor, Codex and Replit, the developer increasingly starts with a goal. They describe what needs to be built, provide context and constraints, and let the system generate, inspect, test and iterate.

The important change is not simply that AI writes more code.

It is that the human is moving from doing the task to orchestrating the outcome.

Instead of thinking primarily about individual lines of code, the developer starts to think at a higher level. What are we trying to achieve? What constraints matter? What should the system attempt? Where should I intervene? Is the output good enough?

That is a very different mode of work.

You can think of the shift as moving from tool to copilot to harness to orchestration.

A tool helps you do the work. A copilot assists while you work. A harness gives AI the context, permissions and tools to act. Orchestration is where the human starts directing that system toward a goal.

This is why software matters so much as an early signal.

It shows that AI disruption is not just about replacing one task with automation. It is about moving the human up the abstraction layer.

When that happens, the worker spends less time manipulating the underlying artifact and more time setting goals, reviewing outputs, managing exceptions and making judgment calls.

That is probably the pattern we should expect to see elsewhere.

A lawyer may move from drafting documents to orchestrating a matter. An accountant may move from preparing accounts to supervising the financial close. A recruiter may move from searching databases to orchestrating a hiring outcome.

Software is simply the first place where we can see that transition clearly.

From Tools to Outcomes: The Canva Example

Design gives us another useful way to see the same pattern.

In the early days, creating a professional piece of visual communication required specialist skills and specialist tools. You might take a photograph, edit it in Photoshop, lay it out in QuarkXPress, prepare it for print, and then publish it.

The work was fragmented. The tools were powerful, but they assumed expertise.

Then Canva came along and changed the equation. It did not eliminate design, but it dramatically lowered the skill threshold. Templates, drag-and-drop editing and browser-based publishing made it possible for almost anyone to create something that looked professional.

That was a major act of democratisation.

But AI raises the abstraction layer again.

Instead of opening a template and designing a launch poster, the user can increasingly say something like: “Create a beautiful launch poster for this product and make it appealing to these customers.”

The user is no longer thinking primarily about layout, typography, image treatment or export settings. They are describing the outcome they want.

That is the same pattern we saw in software.

The progression moves from expert operation, to simplified creation, to outcome orchestration.

And this creates an interesting strategic question for Canva itself.

Canva disrupted tools like Photoshop by making design easier. But if AI keeps moving the interface upward, Canva has to keep moving too.

The next layer is not just “make a poster.” It is about building a brand, launching a brand, growing a brand.

That means moving beyond the asset and owning more of the surrounding workflow: positioning, identity, campaign creation, publishing, optimisation and distribution.

If Canva can do that, it moves from being a design tool to being an outcome platform.

If it cannot, another AI system may sit above it, own the user’s goal, and use Canva or similar products as invisible infrastructure underneath.

That gives us another important pattern:

Tools → Workflows → Outcomes

As AI raises the abstraction layer, value may increasingly move toward the system that understands and owns the user’s goal, rather than the tool that performs one part of the task.

Where AI Disruption Is Showing Up Next

If software gives us the clearest picture of where AI disruption can go, the next question is where we can already see similar signals outside software.

The strongest early example is probably customer service. AI is moving beyond helping agents find answers and into handling conversations directly. In some environments, the human role is already shifting toward supervision, escalation and exception handling.

Marketing and creative production are also moving quickly. AI can now produce copy, imagery, campaign variants and large volumes of content at very low marginal cost. The more interesting shift is not just faster production, but the beginning of end-to-end orchestration across creative, media, publishing and optimisation.

Translation and localisation have already gone through a similar change. In many cases, humans increasingly review and refine machine-produced output rather than creating the first draft from scratch. That clearly signals the worker moving from producer to supervisor.

Accounting looks highly exposed because so much of the work is structured, repetitive and rules-based. Reconciliation, categorisation, reporting, anomaly detection and month-end processes are all natural candidates for AI. The technology is already assisting heavily, but broad execution and organisational redesign are still emerging.

Legal is similar. AI can already support research, contract review, discovery, drafting and due diligence. But the cost of errors is higher, and questions of liability, trust and professional responsibility slow the move from assistance to execution.

Recruitment is another strong candidate. Sourcing, matching, outreach, screening and scheduling can all be automated to a meaningful degree. But persuasion, trust, culture and final hiring judgement still keep humans deeply involved in the process.

So the pattern is uneven, but visible.

The important signal is not how many people are using AI. It is whether the human is still doing the underlying work.

When people start spending more time setting goals, reviewing outputs, handling exceptions, and supervising AI systems, the shape of the work begins to change.

That is when adoption starts to look like disruption.

Why AI Disruption May Take Longer Than the Hype Suggests

This is where history becomes useful.

Major technologies often arrive long before their full economic impact becomes visible. The invention appears first. Productivity gains, organisational changes, and new business models come later.

The PC is a good example. Computers clearly changed what was possible, but companies still had to redesign processes, retrain people, connect systems and rethink how work should be organised. The technology moved faster than the institutions around it.

AI looks similar.

The models are improving quickly, but companies do not transform at model speed. They have legacy systems, regulation, budgets, incentives, existing teams, customer expectations and internal politics. Even when the technology can do the work, the organisation may not be ready to hand it over.

That creates a gap between technical capability and economic transformation.

This is also why the current debate can feel so confusing. AI investors can look at the capability curve and see enormous disruption ahead. Employees can look at their daily work and see that surprisingly little has changed. Both observations can be true at the same time.

The bottleneck may no longer be the model's intelligence. It may be the speed at which organisations can redesign themselves around it.

There is also a trust problem.

Software has an advantage because much of the output can be tested. Code can run or fail. Tests can pass or break. Errors can often be caught before the product reaches the customer.

That is harder in legal, accounting, healthcare, recruitment and other professional domains. The output may be plausible but wrong. Mistakes can have significant consequences. Someone still needs to own the judgement and the liability.

So even if AI becomes technically capable of doing more of the work, the move from produce to execute may be much slower in high-trust, high-consequence environments.

This is why I am cautious about some of the most aggressive forecasts.

I think AI has the characteristics of a major disruptive technology. I also think we may underestimate how long that disruption takes to move from software into the wider economy.

The capability can arrive quickly.

The transformation usually does not.

Signals That Real AI Transformation Is Underway

If adoption is not the same as disruption, then we need better signals than usage numbers.

The first signal is when humans stop creating the first draft. Instead of producing the work and asking AI to improve it, the AI produces the work and the human reviews, edits or approves it. That is already visible in software, translation and parts of marketing.

The second signal is when interfaces start disappearing. Instead of opening several tools and completing a sequence of steps, the user expresses a goal in natural language and the system decides how to get there. This is the move from operating software to orchestrating a workflow.

The third signal is when AI starts taking actions rather than making suggestions. Drafting an email is assistance. Sending it, updating the CRM, scheduling the follow-up and changing the next action based on the response is execution.

Another signal is when several tools become coordinated through a single harness. The user no longer thinks about which application does which task. They think about the outcome, while the system decides which tools, data and actions are required.

Then the economics begin to move.

Pricing starts shifting away from hours, seats and manual effort toward outputs and outcomes. If a legal task takes ten minutes instead of ten hours, the hourly billing model becomes increasingly difficult to defend. The same is true in recruitment, accounting, agencies and many other professional services.

Headcount is another important signal, but probably a later one. The strongest evidence of disruption will not be a company announcing that it has “adopted AI.” It will be a company redesigning a workflow so that fewer people are required to produce the same or better outcome.

That is the point where the technology starts changing the organisation rather than simply helping the organisation.

So the signals I would watch are not model benchmarks or chatbot usage. I would watch for humans becoming supervisors, prompts replacing interfaces, agents taking actions, workflows collapsing into orchestration, pricing models changing, and organisational structures following behind.

That is when AI moves from being impressive technology to becoming economic disruption.

What Happens Next

If this pattern is right, the next phase of AI will not be defined by better chatbots. It will be defined by systems that can take a goal, understand the context, use the right tools, complete more of the work, and improve from the result.

That means the centre of gravity keeps moving upward. We move from tools, to workflows, to outcomes. For workers, that means less time manipulating the underlying artifact and more time setting direction, defining constraints, reviewing outputs, handling exceptions and making judgement calls.

For companies, that creates a much bigger strategic question. Do you own the tool, the workflow, or the outcome? As AI becomes more capable, the value of the lower layers may compress. If another system understands the customer’s goal and can orchestrate the work end to end, the tools underneath can become interchangeable.

That is the risk for many software companies today. A product that was once the destination can become a feature inside someone else’s workflow. But it is also the opportunity. Companies that move up the stack can become much more important to customers because they stop selling access to software and start helping customers achieve something they actually care about.

For Canva, that could mean moving from helping people make assets to helping them build, launch and grow a brand. For recruitment, it could mean moving from sourcing candidates to delivering a successful hire. For accounting, it could mean moving from bookkeeping and reporting toward running the financial close and surfacing the decisions that matter. For legal, it could mean moving from document production toward managing a matter through to an outcome.

This is why I think the next few years will be less about whether AI gets adopted and more about where ownership moves. The technology is already proving that it can assist. In some areas, it can now produce. In a smaller number of cases, it can execute. The bigger transformation comes when organisations redesign themselves around those capabilities.

That process will probably be slower than the hype suggests. Companies have legacy systems, regulation, trust issues, incentives, existing teams and customer expectations to work through. But history suggests that once a general-purpose technology becomes embedded deeply enough, the organisational changes eventually follow.

So I do think AI could become one of the major disruptive technologies of our time. I just do not think that means every industry changes next year.

The more useful question is whether we can recognise the shift when it starts.

When humans move from producing to supervising, when prompts replace interfaces, when systems take actions rather than offer suggestions, and when companies start selling outcomes rather than access to tools, we will know we are moving into the next phase.

That is when AI stops being impressive technology and starts becoming real economic disruption.