AI is a Competitive Advantage

Written by Sam Hudson, with research and editing assistance from AI

In my last article, I argued that code complexity is becoming a less reliable moat. AI has reduced the cost of building software, and that gives businesses and consumers more options than they've had before. Replacing a vendor, building it yourself, or negotiating a better deal.

This article will aim to focus more on the "why." Why are some moats eroding while others remain?

I think the answer comes down to what's actually protecting them. If the only thing protecting a moat was that intelligence was scarce and expensive, AI is removing that protection. If it's protected by something else, it's probably more durable.

Limiting Factors of Intelligence

Dario Amodei (CEO of Anthropic) has a framework that I think helps make sense of this. In Machines of Loving Grace and The Adolescence of Technology, he argues that intelligence is only one input into progress.¹ Even as AI makes intelligence more abundant and accessible, five other factors remain as bottlenecks:

Speed of the outside world. Physical things take time. Manufacturing, experiments, institutional change. You can't think your way past this.
Need for data. If the data doesn't exist, more intelligence doesn't help. You have to go collect it, and that takes time too.
Intrinsic complexity. Some systems are chaotic. Human behaviour, brand perception, social dynamics. Intelligence only gets you marginally better predictions in these systems.
Constraints from humans. Regulation, law, habits, willingness to change. Plenty of things that work technically are held back by human factors. Amodei cites nuclear power and supersonic flight as examples.
Physical laws. Hard ceilings. These don't move.

These five factors are not about intelligence. They're about everything else. As AI reduces the intelligence bottleneck, these remain regardless of how capable the models get.

This suggests a useful way to think about any competitive advantage: what is actually protecting it? If the main thing protecting it was that intelligence was scarce and expensive (it was hard to build, it required rare talent, it took clever engineering), that protection may weaken over time. If it's protected by one of Amodei's five limiting factors, it's probably more durable.

Coding is where this shows up first, because it's one of the few valuable economic activities that isn't really constrained by any of the other five bottlenecks. No physical processes to wait for, no regulatory gates, no dependence on scarce real-world data. Nearly pure intelligence work. Which is probably why AI has improved at it so fast, and why software is worth looking at as an early indicator for other industries.

Competitive Advantage

Hamilton Helmer's 7 Powers is the standard framework for durable competitive advantage.² Scale economies, network economies, counter-positioning, switching costs, branding, cornered resource, and process power.

When you hold these up against Amodei's limiting factors, a pattern shows up. The barriers that depend primarily on intelligence scarcity (expensive R&D, complex integrations, large engineering teams) are the ones under most pressure. AI reduces the cost of R&D, writes migration scripts in hours, and increases leverage per engineer to the point where headcount matters less.

The barriers protected by the other five factors look more durable. Brand trust is protected by intrinsic complexity. Regulatory approvals, contractual lock-in, and organisational habits are all protected by constraints from humans. Proprietary data is protected by the need for data and speed of the outside world. You can't synthesise ten years of real customer behaviour, and SOC 2 compliance takes years regardless of how capable the AI is.

Counter-positioning is an interesting exception. It may actually be strengthened by AI. New entrants can build competitive alternatives faster than ever. But the incumbent's barriers to responding (revenue considerations, board expectations, organisational identity) are all constraints from humans. The offence gets faster while the defence stays roughly the same speed.

The pattern that starts to emerge isn't really about individual powers holding up or not. It's about who is positioned to take advantage of the shifts across all of them.

And that might be the more interesting question. Not "which powers hold up?" but "who is best at using AI to take advantage of the changes happening across all of them?"

A company that can use AI to move faster, build alternatives quickly, accumulate data advantages, and deliver quality at speed isn't relying on any single power. It's working across the shifts between them.

Implications Beyond Silicon Valley

If organisational AI capability does turn out to be a meaningful advantage, it's worth noting that it doesn't belong exclusively to Silicon Valley firms or startups.

Any organisation that develops strong AI practices could start building its own tools where it makes sense. Not replacing every SaaS subscription overnight. But having the option to, in specific areas, and knowing that the option exists.

That's part of what makes option B (build it yourself) from Part 1 worth considering. Even if a company never fully exercises that option, being able to changes the conversation.

Enterprise software is deeply embedded, and there are good reasons (reliability, security, support, integration depth) why organisations pay for established products. None of that is going away. But the range of options available to buyers is gradually widening.

Looking Ahead

In the long run, whether it's a new entrant, an incumbent, a supplier, or a customer, the companies that treat AI as an opportunity will outperform those that treat it as a threat to manage.

By "treat it as an opportunity," I don't just mean adopting AI tools. I mean understanding how AI changes the competitive landscape around your business, and being agile enough to respond or disrupt when it arrives. Right now, software development is the clearest example. But this dynamic will likely reach other industries as AI capability improves, just at different speeds depending on how much Amodei's limiting factors constrain them.

Every situation is different, but Helmer's 7 Powers provides a useful starting point for thinking through which of your moats are protected by something durable, and which ones were only ever protected by the scarcity of intelligence.

For SaaS companies, that question is already urgent. Understanding which of your moats actually hold up, and shifting toward the ones that do, is not something that can wait. For companies in other industries, there's more time, but the pattern is the same one we've seen with every major technology shift. The organisations that embraced the internet early didn't just survive the transition, they shaped it. AI is likely no different.

Sources:

Dario Amodei, Machines of Loving Grace (October 2024) and The Adolescence of Technology (January 2026)
Hamilton Helmer, 7 Powers: The Foundations of Business Strategy

Limiting Factors of Intelligence

Competitive Advantage

Implications Beyond Silicon Valley

Looking Ahead

Sources:

More interesting reading: