The Questions Stack Overflow Now Rewards

I have been on Stack Overflow long enough that certain reflexes are burned in. One of them fires when a question asks what the best practice is, or which tool to use: the old training says flag it, vote to close, link the asker to the page explaining why we do not do opinions here. So it stopped me cold, browsing recently, to see live questions wearing the tag best-practices, undeleted, unclosed, gathering answers in the open. Not one stray question either; the tag sits there with a growing pile under it, next to a sibling called tooling-recommendation that names, in plain words, the exact category the site spent my entire career closing on sight.

To feel why that is remarkable you need the history, because these are not just tags. They are the ghosts of a constitutional purge.

A dead tag walking

In August 2010, Stack Overflow published a post called The Death of Meta Tags and systematically destroyed the tags that described what a question was rather than what it was about. The most famous casualty was best-practices. The reasoning was clean: best-practices questions attract opinions, opinions cannot be verified, and the site was built on the premise that a programming question has a checkable answer. The same logic later pushed every "which tool should I use" question off the site entirely, to the point that an exile colony was founded for them at Software Recommendations, a separate Stack Exchange site whose main function, as far as I could ever tell, was to be the place Stack Overflow pointed at while closing your question.

That was the constitution for fifteen years. Facts in, judgment out.

The reversal, it turns out, is official. In October 2025 the site quietly began testing what it calls open-ended questions, the developer-preference, more-than-one-right-answer kind it had always closed, and by February 2026 the experiment was expanded to everyone. The resurrected tags are the visible edge of that: the machinery of the old ban being dismantled piece by piece, in production.

Amending a constitution in public

What makes this more interesting than a feature announcement is watching the community metabolize it, because the debate is happening in the open, on Meta, and it reads like a constitutional convention.

There is a thread asking, earnestly, what should even be tagged best-practices, which cites the 2010 execution the way lawyers cite overturned precedent. The community that once burninated the tag, with tooling built specifically for burning tags at scale, is now drafting its usage guidance. Another asks whether open-ended questions should be held to different standards, whether the site can afford to be more lenient with them, which is the whole fifteen-year argument compressed into one title. In neighbouring threads the amendments pile up: should old questions, closed as opinion-based under the previous regime, be retagged and reopened, like convictions revisited under a new law? How should the Staging Ground, where new questions are coached before they go live, handle one that has no right answer by design? Should an open-ended question be required to carry a second tag, so it declares itself honestly?

None of this is unanimous, and calling it a warm welcome would be a lie. The official posts laying out next steps for open-ended questions have been voted down hard, while the community's replies to them are among the most upvoted things on Meta, which is the site's traditional way of saying we heard the plan and we have notes. That argument is exactly what makes the shift credible. A top-down rebrand would be easy to dismiss. A community fifteen years deep in one doctrine, fighting clause by clause over how to operate under its opposite, is something rarer: an institution changing its mind in writing, with the objections preserved in the record.

The votes agree

If the tags and the threads were the whole story you could dismiss them as policy theater. What convinced me is that the votes are moving the same direction, and votes are the one thing on Stack Overflow nobody can decree.

One of the most read programming questions on the site recently was not a segfault or an off-by-one. It was a student writing, in a title with no code in it at all, "I get stuck learning HTML, CSS, and JavaScript (I am a student)". No error message, no minimal reproducible example, no single checkable answer. By the old constitution that question is unsalvageable, too broad, primarily opinion-based, closed in minutes with a link explaining how to ask. Instead it collected thousands of views, a question score of 19, and a top answer at 17 in which a stranger walks a beginner through how to be stuck without drowning. I have argued in an earlier essay that being stuck is where understanding forms, so I will admit the question found a soft target in me. The voters did not know that, and it did not matter.

The pattern repeats wherever I look lately. Someone posted their inventory management project and asked, in painfully honest English, for people to critique it and tell them how to get better; the kind of open-ended request that used to be the canonical example of what Stack Overflow is not for, and its top answer sits at a score most answers on the site never reach. "Is coding going to be obsolete?", a question that is one hundred percent career anxiety and zero percent code, quietly gathered hundreds of views. Meanwhile the fact-shaped questions, the how-do-I and the homework and the paste-my-error, still land at the bottom, downvoted and closed, not because they got worse but because a machine answers them before the tab finishes loading, and everyone, asker and voter alike, knows it.

Why now

None of this is mysterious if you follow where the value went. I wrote in an earlier essay that AI makes the cheap half of knowledge worthless and the expensive half more valuable than ever. Stack Overflow's rules were an engine for manufacturing the cheap half at industrial scale and archival quality: canonical, deduplicated, verifiable answers to questions of fact. It succeeded so completely that its archive became training data, and the machines trained on it now answer those questions faster than the site ever could, without the tone. The objective middle of programming Q&A, the how-do-I and the why-does-this-error, has been automated out of scarcity.

What is left scarce between people is exactly what the constitution excluded. Which practice is sound. Which tool is worth adopting, from someone who adopted it and paid for the mistake. Whether this code, my code, the one project I care about, is any good, and what specifically to do about it. Judgment, in a word, and judgment was banned precisely because it cannot be verified, only trusted. The site optimized itself, with real rigor and the best intentions, for the one category of content AI would automate first, and is now, officially, reopening its doors to the survivors.

There is a version of this observation that gloats, and I want no part of it. The strict years were not a mistake. The rules produced the finest programming reference the world has, and the discipline of asking a question the Stack Overflow way, minimal example, exact error, what you tried, taught a generation of us to think before asking, myself included. The constitution was right for an era when facts were scarce and had to be mined from people. The era ended. The credit due to the site is that it noticed, and the credit due to the community is that it is doing the hard part, renegotiating fifteen years of jurisprudence thread by thread, rather than pretending nothing changed.

What the amendment admits

Step back far enough and the tags, the threads, and the votes are all one signal. The reversal is a confession, made in policy form: the part of the archive that made Stack Overflow indispensable is no longer the part machines cannot replace. Facts are handled. What the site is betting its second act on is the question type it once considered beneath it, the unverifiable, the situated, the judgment call, brought to people because only people have paid the price of holding an opinion.

My closing reflex, trained by the old constitution, still fires when I see a best-practices tag. I am learning to let it pass. If the counts keep climbing, then somewhere between a resurrected tag and a Meta thread this year, an era ended, and the site that taught the world programming facts began turning into the place you bring the questions machines cannot settle. That is a smaller site, almost certainly. It might also be the more valuable one.