In his latest blog post, Salvatore Sanfilippo argues that coding assistants are a great democratizing technology, provided that they are not solely in the hands of a few big companies.
I can't claim anything resembling his credentials in software development, but I think I do see the outline of why this will be controlled by a few players who will have a majority of the market share.
And part of the reason lies in the nearly complete collapse of Stack Overflow. Yes, the atmosphere wasn't great. But people didn't suddenly stop running into programming problems. There are other sources of help (irc, Discord, mailing lists) that can be more welcoming and knowledgeable, but Stack Overflow is still the most visible source of aid on the public internet.
People have simply started directing their questions towards LLMs and coding assistants.
And those questions are beautiful training data for those same models. Everything that used to be asked in a public forum, visible to everyone, is now a private confession to an LLM that tells you you're smart for asking such insightful questions. They'll never judge you, they'll never tell you that you're holding it wrong, because it's the questions you ask that feed them.
The most popular LLMs will experience a feedback loop. More people will use them, which means that they will have more access to up-to-date training data, which means that the quality of their answers will be better.
Meanwhile, the disappearance of public forums means that publicly available information will get more stale, and less comprehensive.
I'm just old enough to remember programming before internet access became ubiquitous. It wasn't that different, sometimes you got stuck, and you had to force a solution to a problem by stupid experimentation. But when you program for money, getting stuck on a problem thousands of people have encountered before is wasteful and expensive. For all of my professional life, I've been able to Google the meaning of obscure compiler errors and weird API behavior, and I had a good chance of finding a solution because other people had encountered them before.
When the public forums go silent, answers to many programming problems will no longer be a Google search away, but LLMs will know them. Your employer will require you to get an LLM subscription. Your employer will probably pay for that subscription, and salaries will adjust to compensate.
The word “enclosure” refers to a historical development in England, where common grounds were turned into privately-owned plots. Stack Overflow, despite its many flaws, despite being privately owned, mostly met the criteria for being a public good. This public good, our own willingness to share our ignorance and knowledge in a public space, is now redirected towards private interests.
Pundit Yanis Varoufakis has talked about “cloud rents”, money earned by cloud companies simply by their near-monopolist ability to mine user interaction for data. When this user interaction was previously available without cost to all, I think the word “rent” is particularly apt.
I've talked about this with some people. While to me this seems obvious, others have said that there's no reason to assume it will play out like this. Some people are worried about coding agents being able to work completely autonomously; I think this is immaterial compared to the effect of data piling up in the knowledge silos. So as of January 2026, this is not entirely obvious. Maybe the above will seem obvious in a few years, or maybe this will be an embarrassing thing to read back.
There may be an obvious way to resist this development, but I don't see it. If you keep using Stack Overflow, that data will be ingested for training by LLMs. Any kind of strategy based on openly sharing information will maintain the asymmetrical advantage enjoyed by the big corporate players.
But not sharing information in the open will increase their advantage even more. It will handicap programmers who forego LLMs. It will handicap people who build open source LLMs. All while increasing numbers of people pay for the privilege of increasing these private data hoards.
Let me append some additional predictions: These feedback loops will mean that some LLMs get better at some subjects than others. Microsoft Copilot will be better at solving problems with Microsoft technology, not just because it has all the internal documentation, but because Microsoft users are far more likely to use Copilot, so Copilot will be aware of the problems Microsoft users actually run in to. Some of these feedback effects will occur spontaneously; Claude may be slightly better at (for example) Ruby programming than other LLMs, which will cause Ruby users to rely on it more, which will eventually make Claude the only reasonable option for Ruby programmers, etc.
Another obvious one is pricing. Developer salaries differ by an order of magnitude depending on where you live in the world. $300 a month may be negligible to someone working in Silicon Valley, but may be enough to push a programmer in a developing country below the poverty line. The big LLM-platforms will have to balance their profit margins with their desire to harvest feedback from as many people as possible. My prediction is that they will introduce complicated and regional pricing structures to extract from each according to their ability to pay.
Companies will also have a unique opportunity to build their own, private knowledge silos. Instead of training a new generation of junior developers on your ancient, quirky legacy codebase every few years, train an LLM once. Unlike junior programmers, the LLM can't quit in disgust.
I think that employers should be aware that something has been taken from the public. Not just the training data, but the public forums and practices that created this training data in the first place. That at least some of the productivity gains they see from LLMs are not due to their "intelligence", but from information they used to get for free. And that LLM companies are now selling back to us something that used to be available for free.