Andy Matuschak has a particularly insightful note about how novel user interface ideas are public goods:

One core challenge for Tools for thought is that their core insights are generally expensive to produce but cheap to duplicate. For example, Adobe developed countless interactions and metaphors in Photoshop and Illustrator which Sketch and Figma could copy and improve without paying for years of R&D. Apple spent years developing modern multi-touch interfaces; Google copied them in under a year and suffered little in the ensuing lawsuits. A novel interface idea is roughly a Public good: they’re non-excludable (Excludability), modulo the relatively smaller costs for a competitor to duplicate the idea. Happily (for society), such ideas are non-rivalrous (Rivalry) or even anti-rivalrous.

The last part is a bit hard to parse with its economics terminology, here’s my simpleton english translation:

When someone invents a novel interface idea, anyone can copy it (non-excludable). One company using the idea doesn’t stop others from using it too (non-rivalrous), and it actually gets better for users when many products use the same idea (anti-rivalrous).

Adobe had to invent a huge number of the interaction patterns that now feel obvious in Photoshop and Illustrator. But once those patterns existed, competitors could imitate them at a fraction of the cost. Apple spent years developing modern multi-touch interfaces; others adopted the same basic ideas much faster afterward. Once someone has shown the way, the second mover’s job is easier.

What’s striking in 2026 is how much bigger this phenomenon has become. It’s no longer just interface ideas that are public goods. Increasingly, software itself is.

There was a time when copying software was hard and somewhat slow. Even if a competitor could see what your product did, reimplementing it required time, labour and skill. That lag created a kind of moat that mattered somewhat, but now it no longer exists.

Cloudflare rebuilt Next.js in one week, Naval says “pure software is becoming uninvestable” and Jack Dorsey cut 4000 jobs at Block because “intelligence tools have changed what it means to build and run a company.”

You can object that these are edge cases. And they are. Rebuilding one framework is not the same as building a durable business. A tweet is not a theorem. A CEO justifying layoffs may be partly responding to markets, not merely technology. But the direction still seems hard to miss: the cost of reproducing digital products is falling fast. We’re rapidly approaching a point where a small team can point a swarm of Claude agents at an existing project and have a working reverse-engineered app up and running in no time.

Up to this point, writing software has been an arcane craft practiced by highly specialized artisans. Software was hard to make in the way shipbuilding or watchmaking is hard: not merely because it required labor, but because it required a rare kind of skill. Very few people can do it, and even fewer people can do it well. There were a few consequences of this:

1. There’s always infinitely more work to do than software engineers to do it. Creating an insatiable demand for software engineers.
2. Producing software was always much more expensive than it first appears, even when taking this rule into account.
3. Software talent is a real thing that’s very hard to quantify, impossible to credential, and extremely in-demand. This idea is popularized by the idea of 10x engineers, but it’d probably be more accurate to talk about 100x engineers.

What AI changes is not just the speed of programming, but the economics of it. If you can get much of the production work done by machines, then the old bottleneck disappears. The supply of software stops being tied so tightly to the supply of skilled programmers. That has several immediate consequences:

1. Software production will be rapidly automated, and it’s already happening.
2. It’s possible to produce software dramatically cheaper and faster.
3. It’s possible for anyone to produce software without software engineers.

Everything in AI could freeze today and these three facts alone would mean that the entire world is going to change. You can argue about timelines, or about how far the current generation of tools can go. But you don’t need to assume some dramatic science-fiction future to see the change. And even though we don’t know where any of this ends, we know which way the arrow is pointing.

The software industry has hit an industrial revolution / quartz crisis moment and will need to figure out how to compete in a world where everything digital can be trivially replicated. When quartz watches arrived, they didn’t just make watches cheaper, they made the old basis of competition less valuable. Swiss watchmakers survived, but not by winning a head-on fight over accuracy per dollar. They survived by moving the game to branding, status, heritage, emotion, and taste. The product still mattered. But it was no longer enough.

If software can be copied quickly, then software is a weak foundation for a moat. Not worthless. Just weak. The obvious question then becomes: what is worth investing in?

Distribution is one. Brand is another. Proprietary data, trusted relationships, community, regulatory position, network effects, hardware and operational excellence all become more important in a world where features are cheap. The value moves outward, away from the artifact itself and toward everything surrounding it. When R&D on the software itself no longer buys much protection, you have to ask a harder question: what are you building that competitors cannot simply regenerate?

It was the best of times, it was the worst of times, it was the age of wisdom, it was the age of foolishness, it was the epoch of belief, it was the epoch of incredulity, it was the season of light, it was the season of darkness, it was the spring of hope, it was the winter of despair, we had everything before us, we had nothing before us, we were all going direct to heaven, we were all going direct the other way.