Around a year ago I asked whether the cost of building software had just dropped 90%. A lot has happened since then, and I've become more interested in the knock-on question: if building software is that much cheaper, what is the software itself actually worth?

Was software/SaaS a historic economic anomaly?

One thing that has become obvious to me researching this topic is that software is very much an outlier in terms of ability to extract outsized profit margins, and I think this fact underlines why the industry faces such existential challenges presently.

Typically in history you had three main ways of making "outsized" profits:

Firstly, you had monopolistic behaviour, with Standard Oil being the textbook example. While there are no doubt significant market competition concerns in software and SaaS, I would argue most software businesses, at least by number if not revenue, do not operate in a monopoly.

Secondly, you had network effects, restricting new market entrants. Again, the tech giants clearly do have huge network effect benefits, but many smaller players do not, yet still make very healthy profit margins. Newspapers (used to be) a good example of this, advertisers would want a readership base to advertise to, and if you were starting out from scratch with no readership base, you'd very much struggle to make the commercial model work without willingness to take heavy losses while you built up your readership.

Thirdly you had corruption or using violence to extract high profits out of your customers. Protection money from the mafia is the obvious example historically (though still persists in some geographies). Again, there is an argument that many software companies do achieve their market power via regulatory capture, especially in government contracting. And many would argue regulatory capture is corruption with rounded corners. But again, many software companies don't have a lobbying firm working for them, yet still do very well.

While you can argue that some of these factors are in play for certain tech companies, I struggle to see that the average small or mid-tier SaaS or software company takes advantage of this, yet posts usually extremely high gross (and often net) margins.

So why is this?

Intellectual scarcity

I posit instead that these margins weren't driven by these mechanisms, but instead by simply not enough supply of people able to build software for the demand. And while the number of people able to develop software has exploded over the past 20 or 30 years, the amount of software out there has equally exploded.

So in essence, market actors (whether that would be software engineers or companies) could derive a large amount of value for their work, because there is inherently a (fairly) small pool of people able to build said software.

You can see the issue though now AI agents can do a significant proportion of the engineering. That scarcity is evaporating, which in this model therefore drives the value down sharply. The market always beats margins out of you when there is a shift in supply.

Jevons paradox, though?

Of course, it's not quite as simple as AI agents can do coding, therefore the equilibrium between supply and demand changes. Jevons paradox tells us that when something becomes cheaper and more efficient to use, we often end up using more of it, not less. Jevons' original observation was that more efficient steam engines led to Britain burning more coal, not less. So surely with a huge reduction in cost of bespoke software, we'll see far more demand.

This is true - we're seeing an absolute explosion in the amount of software being created. But I'd argue this software is intrinsically different to what made SaaS (especially) so valuable.

The rough idea with SaaS was that you spent some money on development, and then could sell the same software to thousands or millions of customers. Your marginal cost of a new customer was close to zero, at least in terms of software costs.

But with the Jevons paradox-induced software we have now, it is incredibly bespoke for the most part. Why use a generic CRM when you could build (or purchase) one designed for your exact domain?

I think this dynamic explains the real issue here. Yes, there will be software products in the future. But I suspect these software products will be for the most part far smaller in addressable market size.

Baumol's cost disease

Of course, an argument could be that future software companies operate hundreds or thousands of distinct products, across a whole bunch of markets. There is no doubt with agents this could be from a technology point of view at least more possible than before.

But I think the flaw with this strategy is Baumol's cost disease. When productivity shoots up in one part of the economy, the parts that can't speed up end up getting relatively more expensive. The classic example is a string quartet, which needs the same four musicians and the same amount of time as it did in 1800. Apply that to a software business and the bottleneck moves very quickly from building the product to selling it. Already I'm seeing that while building software has got 10x quicker, sales and marketing is as slow as it always was, at least for most B2B deals.

So I think this explains the flaw with the "software factory" concept that many have been pushing. Yes, you can build hundreds of line-of-business applications in a fraction of the time, if you really want to. But selling these to actual humans is just as hard in my opinion, despite the claims of many that they've managed to automate go to market - any (questionable) efficiencies of these approaches are outweighed by the fact that each buyer has far more marketing directed at them.

I therefore think that there is certainly going to be a market for SaaS in the future - not everyone is going to vibe code their own apps, though some will. But my gut tells me that these are going to be far smaller in scale - and therefore the underlying VC model of throw loads of cash at a SaaS business to scale as fast as possible is even more broken than it was before.

Reproduction

The final thing that is really nagging at me is just how easy it is to copy software now. You can take any agent, give it some screenshots (or if you are even more lazy, a browser tab for it to control) and tell it to clone software you think you like.

We've already seen this on an industrial scale with the ArtCraft suite of clones of everything from Excel to Photoshop.

While I'm sure they are not quite as good or stable as the "genuine" product, this has been the biggest shock to me. Arguably already their Photoshop clone has more features than the unfortunately named GIMP open source clone of Photoshop, which has had 30 years of human effort aimed at it.

So this again makes me wonder where the value in software IP lives. Certainly agents aren't (yet) capable of coming up with good software products to build a business around, so there is definitely value in product taste and choosing the right product to build.

But if you do find a successful niche for your product, it is now trivial for anyone to copy that. So this makes me question exactly where the value is going to lie in future.

One other thing I've noticed is that it's much easier to move between software providers now. Agents can export your data, transform it and import it into a new tool very easily. Even if the provider doesn't offer a nice export, agents are very capable of reverse engineering the web app and doing the export for you anyway. Switching costs have always been one of the quieter moats in SaaS, and that moat is draining too.

The writedown

This is where I think it gets uncomfortable for the finance side.

For the past decade or so, the standard defence of loss-making software companies has been that R&D isn't really an expense. It's an investment. You spend heavily on engineering today, and you get a long-lived asset - the product - that keeps paying out for years.

That logic only holds if the R&D actually builds something durable. If a competitor - or your own customer - can rebuild what you spent five years on in a few weeks with agents, then that spend isn't building an asset at all. It's much closer to a running cost you have to keep paying just to stand still. Which means the underlying economics of a lot of software companies are quite a bit worse than the adjusted numbers suggest.

You can see the same thing from the M&A side. A lot of tech acquisitions were really build-vs-buy decisions. Paying 10x revenue for a startup made sense when building the equivalent in-house would take two years and 30 engineers. But if your own team, with agents, can build it in a matter of weeks, the premium for the technology itself drops to close to zero. You're only paying for the customers and the distribution - and if the product has no real traction, the IP on its own is worth roughly what it would cost to rebuild. Which is not very much anymore.

A year ago I was asking whether the cost of building software had dropped 90%. I think the more uncomfortable question now is whether the value of what's already been built has dropped by a similar amount.