Most companies were built around constraints that are starting to disappear.

They were built for a world where information was fragmented, expertise was expensive, systems didn’t talk to each other, and coordination required people. Lots of people. So we built reporting lines, committees, approvals and dashboards. We built entire departments and management cultures around those limits.

But AI attacks those limits, all at once.

That creates an uncomfortable duality many businesses have not yet grasped. Yes, AI is a powerful productivity tool. But that misses the bigger picture. With genuinely smart AI agents emerging, the real question is existential: if we built this company from scratch today, how much of the current structure would still exist?

This kind of first-principles thinking matters for management, but it matters just as much for investors. Without it, inherited complexity can start to look dangerously like competitive advantage.

Safe spaces

The first wave of corporate AI adoption has been framed around productivity, and for good reason. It’s measurable and frankly, obvious – since using an LLM is so user-friendly. It’s also very happy to answer any question within the existing structure.

A company focused on embracing this angle can talk about AI pilots, employee adoption, workflow automation and efficiency targets. This is the normal language of technology adoption in the workplace. It lets everyone agree that AI is important while avoiding the harder, perhaps existential, question.

But as usual, avoiding the difficult question only delays the inevitable.

Company value exists in its outcomes, not its processes. Bank credit isn’t extended because an application happens to move through a certain set of processes or teams, it’s valuable because risk is allocated according to the right balance of judgement, terms and speed.

After an outcome is clearly stated, much of the inherited structure starts to look less like an output engine and more like baggage.

This is the central question AI forces on organisations: are they able to separate the customer outcome from the machinery built around that outcome?

Many companies will say yes, but fewer will act like it.

Moat vs margin

The investor’s job is to find and isolate what makes a company great, what gives it an enduring moat. AI disruption forces a sharper task: distinguishing margin from moat.

Some margins are earned from genuine scarcity: brand, scale, switching costs. And to be clear - AI doesn’t necessarily destroy those. To the contrary, in many cases it could enhance genuine moats.

But other margins are different, they exist because the industry, or the customer’s world, is messy. Put simply; being slow and hard to navigate isn’t a moat, it’s labour with an invoice attached.

AI lowers the cost of navigating friction. It does this by translating between systems, represented by documents, policies and teams. At an abstract, high level, that describes a large share of service work: take messy inputs, interpret them, structure them, and produce a usable output.

The advent of intelligent AI doesn’t mean this work disappears, but it does mean the threshold for what is truly value-add rises dramatically.

If a company’s value proposition is built on judgement, trust, capital, distribution or proprietary access, AI could strengthen the model. But if a company’s value proposition is mostly navigating complexity, the investment case deserves a harder look.

In the near term, it may be hard to see the difference, but in the long term it’s existential.

Think like a programmer

This could sound too abstract, but there’s concrete evidence for what this looks like, if you know where to look. Software development has changed dramatically over the past year, and the past six months have started to redefine what it means to be a software developer.

The early results are dramatic: in some teams, developers no longer write code manually. In their latest earnings report, Spotify’s Co-CEO Gustav Söderström said: “When I speak to my most senior engineers, the best developers we had, they actually say that they have not written a single line of code since December...it’s a big change, it is real, and it’s happening fast”. The human is still in the loop, but their day-to-day job has fundamentally changed forever.

Coding is by no means a perfect analogy for every industry, but it shows the framework clearly. Software has unusually clean feedback loops: tasks can be specified, outputs tested, bugs found. Not only that, but the speed of iteration is extremely fast. That makes it an ideal early case study for agentic AI.

Tools like Claude Code and OpenAI’s Codex aren’t just productivity tools to speed up coding, they actually change the developer’s role. Their tasks become supervision and judgement, which at least for the time being, remain a bastion for human oversight.

It could be tempting to dismiss this as just a software story, but it’s not. Many white-collar processes have a similar underlying structure and objective measurable outputs. The difference is that unlike software, the “code” is not explicit. It’s often buried inside the bureaucracy: job descriptions, committees, templates, spreadsheets and policies. More complicated, sure, but fundamentally only one layer of abstraction away from code.

There is evidence of this outside of software. Performance marketing is one such example. It works, because, like coding, it has fast feedback loops and measurable outputs. This is particularly true for online performance marketing (such as Meta and Google ads) where results are direct – did the ad lead to clicks, or not?

Until recently this ‘loop’ had a well-established framework: a performance marketer had to brief creative, wait for assets, launch variants, check Meta results, kill the losers, scale the winners, then brief the next round. That loop likely measured in days or weeks. With AI agents it can be measured in hours. The economics scale by a similar magnitude.

This doesn’t make brand, taste or strategy irrelevant – far from it. But AI can do the obvious work, things that until recently took a lot of work and coordination. With the commoditized work automated, more time can be spent judging quality, managing risk and deciding which signals matter.

For now, much of this is still clunky and requires technical knowledge. That will change as user interfaces and AI applications get more user-friendly. The important point is that the operating model is seeing a fundamental, and likely one-way breakthrough.

It’s showing us the future, what is possible when work is specified, executed, tested and improved at machine speed rather than committee speed.

The management test

Most executives are trained to see new technology as a productivity tool. With the capabilities of AI that’s no longer sufficient. This is where the contrast between founder-led and inherited organisations is instructive.

Founders start with a customer problem, with the organisation built around that problem. People, systems, governance – all of it gets added over time, but the logic is commercial and direct.

A later-generation operator sees things very differently. They inherit a machine: reporting lines, budgets, committees, risk frameworks, legacy tech, cultural norms and internal stakeholders.

More directly: the founder is laser focused on the product, whereas an operator’s focus can be captivated by the machine.

At its heart, it’s an incentive problem. Founders are rewarded for upside and reinvention. Professional CEOs are generally rewarded for execution, risk management and almost certainly hitting the next budget.

There’s also a cultural dimension. Organisations are not just collections of workflows; they’re groups of people. Headcount, reporting lines and budgets all act as proxies for a very human attribute: power. A CEO may not describe it as empire-building, but the incentives often rhyme with it. Running a larger organisation is synonymous with prestige, influence and, often, more pay.

It’s also important to not be too cynical. Leading a business requires vision, and a belief in growth. Few management teams, or talented employees would naturally be drawn to a company where the strategy was downsizing.

Many people have also built fulfilling, productive careers inside structures that later became obsolete. It’s the role of society and culture to capture and value that. Unfortunately, capitalism is ruthlessly unsentimental.

That matters because for business AI likely forces an awkward question: what if the better business is a smaller business? What if the answer is fewer layers, fewer people? That ought to be great for margins and customers, but it cuts directly against how many organisations understand progress.

Because the current structure has produced profits - and prestige - in the past, it could be mistaken for the asset, especially by those currently at the helm. That’s how incumbent management teams can get the value proposition backwards. But the structure isn’t the asset, it’s the delivery mechanism.

Businesses and investors should know the difference. The companies most likely to succeed probably won’t be the ones with the most Copilot licences, or the boldest AI presentation. They will be the ones willing to rebuild their organisation around the customer proposition.

No doubt, many AI adoption statistics will look impressive. CEOs will talk about the transformative potential for productivity. Investors should look through this noise and focus on harder evidence:

Is cost-to-serve improving? Are products being launched faster? Are incentives changing? Is the customer experience becoming simpler, cheaper or faster?

The path forward

There are some key questions to debate. What does value-add look like in an agentic AI world? One company could become more valuable as complexity falls – enhancing their core value proposition. Another may find their margins depended on complexity staying expensive.

The answers are still murky and will depend on a range of views. Some industries are more protected, some more competitive. Market participants will debate – correctly - about where various players sit in the value chain, and about who really has the moat. But what can be certain is that those debates will be more useful than tracking how often a CEO says “AI” on the latest earnings call.

The best question to ask is: if you started again today would you build the company this way and would it be this big?

For many companies, the honest answer is probably “I don’t know”.

That’s through no fault of their own – most of them didn’t build the business themselves. They haven’t viewed their business through the founder lens. Fortunately, they still have time: they can use it to cut through the structure and ruthlessly focus on what their business does better than any other.

The companies that succeed may be those willing to make their own organisations smaller, nimbler and faster. That is probably uncomfortable for many management teams. But it is the question they, and investors, need to ask. Otherwise, a founder with no legacy structure may come along and answer it for them.

This article was originally published in the NBR on 19 May 2026 (paywalled).  

Talk to us​

If the ups and downs in the markets are keeping you up at night, it may be a good time to check in with a financial adviser and make sure your mix of investments are right for you. You can drop us an email, call us on 0508 347 437, or chat with us online.