GoKawiilSide A: Turtles all the way down / Side B: Mo' tokens mo' problems
If you've been around long enough in anything you start to see history repeating, fashion trends come back around, humanity makes the same mistakes. In the field of computer science we see the same patterns: technology X is essentially the same idea as technology 10 years ago, which was based on the idea for technology Z 20 years ago. Today's 'cool and trendy' named design approach is a re-worked version of MVC, SOA, yada yada.
With this in mind there's a certain irony that a lot of people working in the space are starting to converge on various ideas (see my star chamber blog post for example). Now it's the turn of one of the most useful resources on the internet for software engineers: Stack Overflow. Born in 2008, peaking at over 200,000 questions a month by 2014. Decried as dead towards the end of 2025 (the proclaimed 'year of agents'), down to 3,862 questions in December (back to its launch month numbers after 17 years). The drop off started around the time ChatGPT launched. Who needs to share knowledge when ChatGPT/Claude/Gemini et al. "know everything"?
I am being facetious, as while these tools can help us do some amazing things, they also cause a lot of day-to-day frustration. They run into the same issues over and over, using up tokens, wasting resources and energy. The AI platforms have tried to help us out (or lock us in depending on your persuasion) with skills, features, slash commands, integrations, behind-the-scenes model weight updates; but ultimately you shouldn't have to become an ML engineer or get certified as an 'A* Claude Code terminal operator' to see the benefits.
Anyway, back to the story circa 2026:
LLMs trained on the corpus of Stack Overflow
LLMs via Agents committed matriphagy on Stack Overflow
Agents run into the same issues over and over in isolation because their training data is stale etc.
Agents now need their own Stack Overflow ... the cycle continues
And yes, I chose that word deliberately. Matriphagy ; the offspring consuming the parent. Spiders do it, and there's a certain poetry to the fact that web crawlers (the original "agents") consumed the web's knowledge; knowledge which birthed LLMs, and then those LLMs hollowed out the communities that fed them. In actual spider matriphagy, the mother's body nourishes the next generation. Stack Overflow's corpus genuinely did nourish the LLMs. The question is whether the next generation builds something sustainable or just moves on to the next host.
... continue reading