You can see the question of this essay in the title. It’s something my small support group of fellow programmers and writer/educators message together about regularly. What used to be the provenance of enviously looking at someone who had cracked the code and managed to attract real eyeballs and, therefore, a sales tunnel for a paid course or book, has now become fully existential.

More people than ever are trying to learn how to code, whether through traditional pathways, vibecoding, YouTube videos, or whatever. And yet, at least when it comes to written content, there is almost no SEO path left. Google is no longer interested in showing external links and defaults to AI-generated Gemini answers, while, increasingly, people don’t even use search at all but start with ChatGPT or Claude for such questions.

Look at what’s happened to Stack Overflow, the grandparent of question-and-answer programming sites.

StackOverflow Chart Traffic

The takeaways I see are a serious rise from 2008 to 2014, a leveling off until 2018, a small decline offset by a sharp jump around Covid in 2020, and, since 2022, a dramatic falling off to the point where the site is basically dead. We all know what happened in late 2022 (hello ChatGPT).

This pattern of growth and decline is mirrored in my own metrics and those of most of my content-creator friends. For a brief halcyon phase, roughly 2015-2020, you could write a high-quality, in-depth post on a programming topic and plausibly rank in the Google top 10 for it, receiving solid traffic to then divert elsewhere.

When I look at my 2018 year in review, what jumps out is that I wrote 72! new tutorials, largely because I was learning new things, but also because each one was a chance to rank highly and then be marketed for my books. In that same year, I went from 1,500 weekly unique visitors to 20,000 and was blindly thinking about what would happen if trends persisted.

Book Sales…

These days are different. I still have book sales, mostly through LearnDjango.com and Amazon, though Amazon has undergone a remarkable enshittification process as well that is probably worth its own blog post. But tl;dr Amazon used to use search to show the best results; now it is just like Google, search is completely paid for, and results are about 8/10 ads versus organic links. Oh, and if you pay for ads with Amazon (wink, wink) your organic search results will do better. In short, it is a mess. I used to spend some money on Amazon ads in the early days. It was worth it. I completely stopped 2 years ago because prices spiked 5x and results dropped 3x over the same period.

How is it plausible for any company to buy $3-5 per click ad words on “django book” when they are selling a $40 book, meaning $20 profit, and when only a fraction of people who click buy? The math doesn’t math. At least for me.

And so I still have books on Amazon and provide links via LearnDjango.com, but I don’t focus on that area at all. Because it’s not worth it.

LLM Results

Modern LLMs continue to improve across most benchmarks and in my everyday use. I’m not an AI-luddite at all; in fact, I’ve been using them daily for 3 years now, both ChatGPT and Claude. They have their place for sure. But I will admit I have deep reservations about how long this will persist.

All current models are trained on a free/stolen corpus of human data. Both written tutorials and code. Now, both are dominated by AI. We shall see how good the models really are now that they are consuming themselves.

Then there is the cost structure, where token prices are already rising and need to keep increasing a whole lot more for current frontier galaxy models that burn billions in training and inference to serve.

But for the moment, you can get great results. You don’t need me to tell you how to do something in Django. Just ask Claude to write a custom blog post in the style you prefer on the topic. You’ll get 70% of it right. Yes, some issues, no pretty screenshots, and no guarantee it is, you know, accurate. But better than searching around on the almost-dead internet for info.

The Future

What does all this mean? Maybe AI/LLMs really do figure out how to make software developers and technical writers obsolete. It’s possible. I’m biased against it for a number of reasons, but I would never bet against technological progress entirely.

The real question is: how much value does a human with a reputation and a voice actually add? If you read something from me, or hear me on a podcast, or see me in a video, I’m putting my reputation on the line each and every time. I’m also, hopefully, communicating based on many years of trying to explain topics clearly to people and on my many years as a programmer myself. The same goes for all the other technical content creators out there.

For the time being, I see no point in writing anything not behind a paywall that will just be consumed by an LLM with no attribution. So I’ll continue to write personal pieces, like this, but deep technical dives that take days or weeks? No. Those will be in books, private videos, or in person. It’s not worth it.

Many of my friends aren’t even that lucky. They are planning to just close up shop. No point in updating good materials or even continuing to host what they have. So there goes even more foundational knowledge that the internet and LLMs could have trained on.

I’ll end by making the point that fighting technological change is not a worthy mission. I’m not anti-technology at all. Even if all progress stopped today, LLMs are an incredibly valuable tool that has meaningfully improved my software development. But I harbor real doubts about whether it can continue to improve, suspect it might actually get worse given eroding training data, and I think a whole generation of newcomers is going to be more confused than ever because they are entrusting their educations to a next-token predictor rather than a proper teacher.