Technologists make a common mistake. They often say that tasks humans find effortless, such as movement, perception and common sense, are hard to automate. Meanwhile tasks humans find difficult, such as calculation, optimisation and pattern matching at scale, are easy for machines. This is called Moravec’s Paradox.

Arvind Narayanan[1] questions this belief. He notes it lacks evidence and ignores tasks that are either easy for humans and machines, or hard for both. Narayanan also argues that AI is normal technology. It improves quickly in capability but diffuses slowly through the economy. The gap between what AI can do in theory and what businesses adopt in practice can last decades.

That framing matters because many business owners are drawing the wrong lesson from AI capability demos. They are asking what AI can do, rather than what it makes economic sense for a firm to do.

A simple way to think about this is with a 2x2 grid. One axis is easy for humans versus hard for humans. The other is easy to automate versus hard to automate.

Start with the quadrant that is easy for humans and easy for AI. This includes data entry, basic classification, scheduling, standard reporting, and routine customer queries. These tasks will be automated aggressively because the economics are obvious. They are repetitive, low risk, and cheap to replace. If your business model depends on charging for this work, your margins will compress. If your staff spend most of their time here, you are wasting expensive human capacity.

The second quadrant is easy for humans but hard for AI. This is where many small professional firms believe they are safe. Relationship management, strategic consultancy and messy real-world coordination all live here. These tasks will remain with humans, but AI may put pressure on margins.

If something is easy for humans, it is usually easy to train, easy to copy, and hard to defend. AI does not need to fully automate these tasks to weaken them economically. It only needs to lower the skill threshold or reduce the time required. That increases supply and pushes prices down. Many knowledge workers are about to discover that being “not yet automatable” is not the same as being valuable.

The third quadrant is easy for AI and hard for humans. This is the most interesting and the most neglected. Large-scale optimisation, rapid scenario testing, and complex forecasting are cognitively exhausting for people but increasingly tractable for machines. Solve problems in this quadrant and you will grow revenues for you and your clients.

A small firm can price more accurately with predictive analytics and dynamic pricing. It will respond faster to demand changes, identify profitable niches earlier, and personalise offers at scale. Developing products that do these things for clients creates further opportunity. Smart automation focuses firms on things competitors can only do by blowing up headcount and cost.

This is where Narayanan’s point about diffusion matters. These advantages take time. They require integration into workflows, trust in outputs, and changes to decision-making authority. That slows adoption but builds a moat for first movers. Once embedded, advantages are sticky. Firms that learn to operate in the third quadrant will pull away from those stuck automating admin.

The final quadrant is hard for humans and hard for AI. This is where small businesses go to die. Overly complex offerings, bespoke processes that cannot be standardised, and services that require constant exception handling fall into this category. They are expensive to deliver, hard to explain to customers, and impossible to scale. Spending time in this quadrant exposes the business model.

Many founders are attracted to this quadrant because it feels sophisticated. In reality, it is fragile. When margins tighten or demand shifts, these businesses have no room to manoeuvre.

Narayanan’s core argument is that AI should be treated like electricity or the internet. Transformative but far from instantaneous. The firms that benefit the most will be those that align technology with economic incentives and organisational reality.

That brings us to the business lesson that matters whether AI lives up to the hype or not. Good businesses do things that are easy for customers to adopt. They reduce friction while saving time and money. Or they help customers make more money with less effort.

AI can support that, but it does not change the rule. If your product requires customers to radically change behaviour, trust opaque systems, or tolerate frequent errors, adoption will be slow. Normal technology diffuses at the speed of institutions, not demos.

The opportunity for small businesses is to earn revenue from work that machines are good at and humans are not. Do not rely on selling automations of tasks that machines and humans both do cheaply, but do adopt them in-house. That is how you protect margins and create leverage.

Moravec’s Paradox is a map of economic pressure. It’s a simple way to think about where best to position a business. Those that choose the wrong side of easy get squeezed from both directions.

[1] Fact Checking Moravec's Paradox