June 10, 2026
Not too long ago I went to a fireside between industry veterans with opinions about devtools and open source. During the conversation, one of the two posed a thought to the audience to consider a world in which “software is commoditized”. When you call a car on Uber, you only care where the car is; so what if you could generate a UI on the fly showing you just the location and ETA instead of what we have today with all the buttons and distractions? Mind you, this was from the same person who, in response to a question about what happens when VC capital stops subsidizing tokens, chortled that people will pay nonetheless. At the time, it gave off more of an impression of “I’m wealthy and enjoying it” (he works at a big name firm), but with what coding agents can do today plus what they’d be able to accomplish down the line, it doesn’t seem so far stretched.
From the burnout of AI psychosis to the ennui from software engineering becoming irrelevant, it’s easy to be someone working in tech or another impacted skill area (not industry) and feel like we can just scroll through websites engineered to our dopamine while we sit and wait for UBI (or UHI, take your pick). But, this is the same tune of what we said about self driving cars where, a decade ago, we were told people driving would become meaningless in five years.
What if I told you the commodification of software is an inevitable conclusion to a story that’s been building for years? Let’s begin with the opening sentence from geohot’s Sloptember post:
I’m calling it now, the adoption of AI agents into software development will be one of the most costly mistakes in the field’s history.
If you’re skeptical of coding agents, you’d probably agree with the assertion and are waiting for people to stop wasting money on LLMs. If you’re fully surfing the vibes, then you may think he’s “just not using it correctly”. However, let me suggest to you a view that’s not skeptical of coding agents while agreeing with the sentence. The previous event before we commodified software was when we commodified software talent. By this, I don’t mean recruiting but, rather, coding bootcamps.
The sales pitch for coding bootcamps is simple: you spend less time studying than you would on a computer science degree but get the same job in the end. At one point, there was a small frenzy and some would even grant a Bachelor’s degree upon completion. Someone who prided themselves on being one of the nerds who sat in the back of the classroom and studied for years to get ahead of their peers may view this as a saturation of software engineering talent. However, considering software engineers are still people and you’ll have high performers as well as low performers, all it did was add more domestic labor. Coupling this with devshops that just trebuchet engineers at problems (which only saw increased demand through the 2010s as software was eating the world), we were getting closer to a world where more people would be able to code.
If, before, your problem was you couldn’t find a person who knew how to code, now there’s a whole crowd who got the skills you need. However, like why we don’t only hire remote workers for a fraction of the salary, domain expertise can matter as much as high/low performance. Anyone who’s worked on a project with others knows people have strengths and weaknesses; a person who’s awesome at writing SQL queries may be allergic to frontend development whereas the one who’s great with CSS animation transitions may roll their eyes back at the idea of a database. Just throwing money at technical labor without clarity of the goal is as effective as waiting for monkeys on typewriters to hand you Shakespeare.
At the same time, like how all large companies are remote companies, all software practices can be codified and we can assign agents to the responsibilities that comprise an engineering org. Economically, it may be the case that smart people orchestrate systems and spend a fraction of what hiring a team would cost but it may also be the case (as practices are figured out and token prices stop being subsidized) that we end up spending more in total with “artificial talent” than if we had hired a team of people.
As software products continue to become further industrialized, holding opinions such as “LeCunn is right about LLMs” only gets a part of the picture right. Like Warren Buffet’s remark on how he’s “never met a rich economist”, something similar could be said about there not being a directly impactful theorist (emphasizing directly since Marx is what the communists read but Lenin and Trotsky were the ones who directed impact). It is true software engineering is experiencing a deskilling like frontend development, but, it doesn’t change how the internet promises all of humanity’s knowledge yet plenty of dumb people still exist. Just because AI is capable of doing things doesn’t mean it will, someone will still need to invoke it or prepare a system that does so.
Unfortunately, if industrialization and commodification has taught us something (ie mass printing of books, manufacturing and assembly lines), it’s that people will broadly value convenience or accessibility over something more bespoke (why buy a single knife when you can get the whole cutlery set). We could ruminate in that and say this is what people are actually worried about when it comes to AI replacing software engineers. Although, it would be the same as trying to make money with an investment thesis that the world is going to end; buying gold makes no sense and you’re just hoarding bananas. Instead, this should be seen as a dawn similar to that of the internet when everyone online suddenly got the superpower of reading from and communicating with others across the world.
There are numerous problems in math that get shelved or put up with a bounty because the person putting up the problem does not have the capacity or ability to tackle that problem. In the world of software, there have been numerous ideas that wouldn’t exist because there wasn’t any technical effort put in that direction (ie when your friend says “what if there was an app that could X?”). Now, we can technically make anything and delegate our focus more towards the design of the product. In the words of Meek Mill:
I feel like they can make my claude smarter who can help em do that … or what is the smartest ai program available to the people? Because the things I am learning in a week would take me 5 years to learn.