Recently, the venture capitalist Marc Andreesen of a16z discussed the impact of AI on software entrepreneurship, citing Jevon’s Paradox.
Jevon’s Paradox is an economic theory from the 19th century inspired by observations of changing coal consumption.
“Jevons noticed that improved steam engine efficiency resulted in higher coal consumption rather than conservation. He argued that as efficiency of steam engines increased, their usage expanded, offsetting any gains made in energy conservation.”
Steam engines converted coal into power for trains and industrial applications. When steam engines became more efficient at extracting the energy from coal and generating power, we might have expected the consumption of coal to fall. If the demand for power stayed the same and we needed less coal to generate a given amount of power, then it makes sense to think that we would consume less coal.
Instead, when it became cheaper to generate power, the price of power fell. In turn, people used more power than they did before the efficiencies appeared. Supplying the increase in the demand for power required more coal than before the improvement in generation. When it came to coal consumption, the stimulating effect of the lower price of power overwhelmed the conservationist effect of the power generation efficiency.
Andreesen made comments regarding the impact of AI on software development.
The cost of developing software (including starting up a software company) should fall with AI. But there should be a concomitant surge in demand for software capabilities. The price of software will actually rise, in his estimation, because users will be able to do so much more, i.e., the products will be much more sophisticated than what we have today. Customers will be so happy that they will want ever more.
He points to the example of the introduction of CGI technologies in Hollywood. In theory, this should have reduced the cost of production, but in reality the demand for CGI services went up because audience expectations increased so much. Now when you go to the movies, it’s “wall-to-wall CGI” and movies are more “expensive to make than ever.” The result is much more “visually compelling” movies.
In software, products will be “radically” better, in turn driving demand for software.
It’s a good theory. It makes sense. We can imagine it coming to fruition.
It’s interesting in the context of another pronouncement from a16z.
‘“I don’t think it’s crazy to believe that half the white-collar staff at Google probably does no real work,” he said. “The company has spent billions and billions of dollars per year on projects that go nowhere for over a decade, and all that money could have been returned to shareholders who have retirement accounts.”
‘“So those people aren’t just being useless (and being coddled to think useless jobs actually matter – they don’t), but they are also taking money away from the rest of the workforce’s retirement programs,” Ulevitch said.’
What happens when we put these two assertions together?
AI, in making software easier and cheaper to develop, will lead to an explosion in demand for applications with much more attractive feature sets. The price of software will increase to reflect the more sophisticated features we can come to expect. In turn, this means that the Googles of the world will require even fewer employees doing actual work.
What does this have to do with our preoccupation with bureaucracy?
In theory, we should expect to see bootstrapped software companies emerge that have much lower headcounts for an equivalent amount of revenue when compared to even the current best-of-breed software vendors. There should be a corresponding increase to the already healthy margins seen in the industry. The new companies will vomit cash.
But perhaps we shall see something different.
Why is it that companies like Google can carry tens of thousands of employees who are so irrelevant that the elimination of their jobs would have no functional consequences for the level of service the company delivers?
We could speculate that Google hires to provide a margin of safety. They have a deep bench so that the company is invulnerable to the departure of any specific group of employees.
Or we might say that there is a defensive nature to their hiring practices. They bring people inside Google to keep them from working for anyone else.
Possibly, Google management believes in the moonshots. They’ve just been unlucky in converting these projects to profitable businesses. Instead, Google’s best moves have arguably been to acquire other companies that complement and extend the core search and advertising franchise. YouTube is one of the greatest tech acquisitions of all time. Purchasing Android for $50 million was another home run. Google may be better off running their R&D the way Cisco appears to do it: outsourcing to startups that they subsequently roll up if they are successful.
My theory is that Google has the employees that it does for other reasons. They sop up the growing number of elite candidates our universities produce, with the attendant political benefits in Washington DC. Imagine the kind of withering fire Google would have to endure from the “it takes a village to build it” type of regulator if the company had half the employees they currently do, with higher margins and even more free cash flow. It boggles the mind to think of the kind of pushback Google would receive from their customers and the limits this might put on their pricing. Higher free cash flow might make it more likely that new competitors would obtain funding.
We can describe Google many ways. It is not irrational, however. The company may have made a bet that there is an optimal combination of revenue growth and margin beyond which levels they will incur massively punitive regulatory costs, customer pushback, and potential competition.
Call it Sooran’s Paradox: as AI makes things more efficient, it will lead to an equilibrium characterized by even more bureaucracy and window-watching. In addition to making software development easier and cheaper, AI will make it more efficient to add bureaucracy; the technology lends itself to the implementation of rules. Management will tolerate the proliferation of internal regulatory mechanisms to the extent that they don’t impair the core business of the software-consuming customer and they help to optimize margins politically.
If this is correct, we would expect to see relatively higher concentrations of bureaucracy in software-intensive businesses.
Stranger things happen at sea.