In his 2011 article for the WSJ titled Why Software Is Eating the World, Marc Andreesen makes the case that the software industry had reached a level of maturity that allowed it to transform every industry, and in doing so made itself larger, and also caused those other industries to shrink. For example, the mobile phone revolution took a huge chunk of gaming market share away from Sony and Nintendo, and placed it in the hands of anyone with a keyboard and Android Studio. (By the way, one of these little gaming companies went on to become Discord, a company now estimated to be valued at $15 billion.)
But, I think that phase of the software development world is over. We've run out of easily-pickable fruit. The areas of the real world that software is still attempting to transform are big, difficult problems like self-driving cars, VR/AR, and making LLMs resilient enough to prompt injection attacks to make them usable for customer service.
But of course, what's really accelerating this process is Agentic AI coding assists. AIs like GPT5 and Claude Code will significantly reduce the value of software, fundamentally. I have a little side-project I work on from time to time. After several years, I ran scc on the codebase. One of the (somewhat silly and unscientific) things scc does it show you estimates of the amount of effort your project would take to implement:
This is a one-person project that I'm rather proud of, and these metrics do wonders for my ego. But, with AI, I could probably recreate this project in 1/4 the time. And that ratio probably applies to your codebase also. And this is right now. What about in 5 years, when AI context windows grow larger and the cost continues to fall? Software as a commodity simply isn't as valuable as it once was.
Ah, but all of that computing power comes at a cost in the real world, in massive datacenters full of spinning fans and busy techs replacing burned-out parts. In a word, material. Non-software. As software value declines, that value will shift to the providers of the datacenters that enable that decline. And this isn't a one-time setup cost. We will have to maintain massive datacenters for the foreseeable future or the massive cost of software will come roaring back (and it'll be worse, since everyone is addicted to AI coding tools, people who can actually fix things in a reasonable amount of time without AI aids will be rare).
Another important thing to understand is what effect AI will have on tech companies themselves. Companies like Google and Amazon are untouchable thanks to the network effect. But B-tier companies won't be so lucky. If anyone can build an X formerly Twitter (formerly just "Twitter", no "X formerly") clone in a few weeks, what value does the platform have? As former Google CEO Eric Schmidt said in a talk he gave at Stanford University in 2024, "Imagine a [...] programmer that actually does what you want, and you don't have to pay all that money, and there's infinite supply of these programs". At another point in the talk he suggests that if Tiktok is banned (at some future date) that people should just ask their LLM to make a tiktok clone, in hopes that their version will be the one to go viral. The critical point here is that everyone has their own programmer, who will make what they want, eliminating software development from the loop entirely.