A response to Andy Nguyen’s “What AI Did to Stack Overflow, in One Graph That Should Worry Every Developer” – and a follow-up to “The Answer You Can’t Find Again Never Happened.”
A few weeks ago I wrote that an answer you can’t retrieve later might as well never have happened. I was writing about a single learner, the young coder in Tegbi, Ghana or Sokoto, Nigeria who has a breakthrough in a chat window at midnight and can’t find it again by Friday.
This week, Andy Nguyen published a graph that tells the same story at the scale of an entire profession.
His piece traces what happened to Stack Overflow when developers stopped asking the internet and started asking a chatbot. In 2019, your bug led you to a public thread: someone had fought that exact error, posted the fix, and five strangers explained why it worked. Ten years later, the next person with your bug found the answer in seconds. In 2026, your favorite model tells you what to change, you change it, you move on. But then, the answer would have existed for one person, for one session, and then oh no, it’s gone!
Nguyen’s phrase for what we’ve done is blunt: we didn’t just move knowledge around, we buried it. Reading his post, I kept nodding, because he has independently diagnosed the same disease I flinched at in my own build sessions. He just found it in the bloodstream of the whole industry, while I found it in a single learner’s week.
Two symptoms, one disease
Put the two arguments side by side and the pattern is hard to miss.
I argued that for most of human history, the bottleneck in learning was getting the answer. Now, however, with answers infinite and instant, the scarce thing has flipped to keeping them: retrieving, revisiting, connecting. Nguyen shows what that flip does at scale. Stack Overflow wasn’t just a help desk; it was a messy, human-funded encyclopedia of every mistake a programmer can make and how to climb out of it. Community votes sorted right from wrong. Comments caught edge cases. The archive compounded because answers persisted somewhere findable.
The chatbot broke that in exactly the way it breaks the individual learner’s week. Nguyen calls the model “a brilliant tutor who forgets you the second you leave the room.” That is precisely the vending machine I described: useful, fast, forgettable. The difference between a vending machine and a mentor was never intelligence. It was continuity. And continuity is what vanishes when answers live and die inside private, lonely sessions. This is true whether the session involves a senior developer in Seattle or a first-year learner in Gwagwalada.
Nguyen’s most sobering point is about the frontier. The old answers don’t vanish; the 2014 Java thread still ranks. What stops growing is the new record. Every fresh framework enters the world with a near-empty public archive because the people who figure it out do so in a chat window and walk away. His warning for junior developers should ring loudly for anyone thinking about Ghana’s One Million Coders or Nigeria’s 3MTT: strip the archive out from under the next generation, and you don’t get faster programmers. You get programmers who (too often) can’t tell why the code works, only that it worked. Once.
Where we differ, and why the difference matters
Here’s where Nguyen and I part ways, at least at first glance. His fix is public: when the model unblocks you, push the answer back into the light. Write the thread. Add the edge case. Rebuild the commons, ten thousand small acts at a time.
My fix, in the last post, was personal: keep a journal. Build tools, like the AI Learner’s Journal Kit, that we’re putting in front of young coders in Ghana and Nigeria. It saves answers as entries you can reopen, annotate, and reflect on, and it reads your own journal before answering, so the AI remembers your road. Ultimately, our goal is to cultivate astute software developers who can discern why a solution works from a sustainability perspective, rather than mere programmers who just copy what a model spits out
These sound like different prescriptions. They’re actually two doses of the same medicine, and one depends on the other.
Because here is the uncomfortable question Nguyen’s fix runs into: how do you post the answer that unblocked you if, five minutes later, you can’t find it? His entire proposal is to write what you learned and add the edge case the model missed. This assumes you captured the learning in the first place. It assumes you can distinguish the paragraph that mattered from the forty that didn’t, days after the session ended. Anyone who has scrolled hopelessly through an old chat knows that assumption fails constantly. It failed for me this very afternoon, scrolling through a recent Gemini chat, trying to retrieve a paragraph about “basic and derived” potentials, a planning concept I developed in 1987 as a student in the SPRING program at the University of Dortmund, Germany.
You cannot share what you never kept. Private memory is not a retreat from the public commons. It is the seed of it.
The journal as the first act of rebuilding
This is why I’ve stopped thinking of the AI Learner’s Journal Kit as merely a study aid. A journal entry – the answer, in your own words, with the context of what you were stuck on and what finally clicked – is a draft of the public artifact Nguyen wants written. Exactly. The learner who has practiced saving, annotating, and returning to their own answers is the learner who, a year from now, has both the habit and the raw material to write the thread, publish the post, and leave the trail for the next person.
The reflection step matters just as much. We don’t learn from experience; we learn from reflecting on experience. That’s true for the individual, and it turns out to be true for the commons too. What made Stack Overflow valuable was never the first answer, but the reflection layered on top of it: the votes, the corrections, the “this breaks on Python 3.12.” A journal teaches a learner to do privately what the archive once did publicly. Scale that habit across a million coders, and you are not just protecting individual understanding. You are training the generation that rebuilds the frontier’s public record, this time with African learners among its authors, not just its readers.
And the equity stakes I raised last time only sharpen here. A well-resourced developer who loses an answer is inconvenienced; they have colleagues, mentors, slack in their week. A learner juggling a patchy connection and real responsibilities loses the answer and the thread and the momentum. If the public archive thins while personal memory evaporates, the learners with the least slack lose twice.
The quieter race, continued
There is still a loud race for better answers, and the biggest actors can keep running it. Nguyen’s graph provides evidence that the quieter race (for better memory, both personal and collective) is the one that determines who actually thrives.
So, to extend the challenge from my last post: if you’re a learner, journal what you learn, and once a month, turn one entry into something public. If you’re building learning tools, optimize for what happens after the answer: retrieval, reflection, and now sharing. If you’re running a program, measure whether learners can return to, build on, and pass on what they learned.
Because the answer you can’t find again never happened. And the answer you never share dies with your session — no matter how brilliant it was in the moment it arrived.
Building or learning in this space? I’d love to hear how you hold on to – and pass on – what you learn. Find me at agenticppa.com.
Found this useful? Pass it on — that’s rather the point of the whole piece:
Share on LinkedIn · Share on WhatsApp · Share on X · Share on Facebook
