Smith argues that vBulletin/phpBB-era forums didn't lose on merit — Facebook Groups, Reddit, and Discord absorbed the social layer while hosting costs, spam, GDPR compliance, and ad-block collapse made small web properties economically unviable. What replaced forums wasn't better; it was just cheaper for operators and stickier for platforms.
The editorial argues forums had four properties their replacements lack: they were indexed, addressable, threaded, and persistent. Facebook Groups swallowed motorcycle repair knowledge into unsearchable silos, Discord conversations vanish into scrollback, and specific high-signal threads — like a 2007 kernel scheduler discussion that shaped one engineer's career — no longer exist on any archive.
By submitting the essay to HN where it hit 302 points, pentagrama surfaced a thread that became a catalog of specific losses: audio DIY forums with 15 years of tube-amp schematics gone when admins stopped paying hosting, and Something Awful subforums that taught engineers how to reverse-engineer game protocols. The community response frames this as structural knowledge loss, not nostalgia.
The editorial reframes the nostalgia argument as a structural one: the four properties of old forums (stable URLs, Google-indexable titles, linear threaded replies, long-term persistence) are exactly what modern engineering teams need for institutional knowledge. Slack, Discord, and Notion pages fail on at least one of these axes, which is why teams keep re-solving the same problems.
Ernie Smith's Tedium essay — *Bring Back Crappy Forums* — landed on Hacker News at 302 points with a comment thread that reads less like a discussion and more like a group grief session. The premise is simple: from roughly 2000 to 2012, the internet's technical middle class ran on thousands of small, ugly, self-hosted forums. vBulletin, phpBB, Simple Machines, Invision. Ad-supported, often maintained by one obsessive admin, indexed by Google, and full of the kind of specific, weird, high-signal problem-solving that made Stack Overflow's early years possible.
Smith's argument is that those forums didn't die because they were bad. They died because Facebook Groups, Reddit, and eventually Discord absorbed the social layer, while the economics of running a small web property — hosting, moderation, spam, GDPR, ad-block collapse — got progressively worse. What replaced them wasn't better; it was just cheaper for the operator and stickier for the platform.
The HN comment thread turned into a catalog of specific losses: audio DIY forums where 15 years of tube-amp schematics vanished when the admin stopped paying for hosting. Motorcycle repair forums swallowed by Facebook Groups and now unsearchable. The Something Awful subforums that quietly taught a generation of engineers how to reverse-engineer game protocols. One commenter noted their entire career pivoted on a 2007 forum thread about kernel scheduler quirks — a thread that no longer exists on any archive.
The interesting part of this story isn't nostalgia. It's the structural argument buried underneath it, which is directly relevant to how modern engineering teams manage knowledge.
Forums had four properties that the tools that replaced them do not: they were indexed, they were addressable, they were threaded, and they were persistent. A phpBB thread from 2004 had a stable URL, a title that told Google what it was about, a linear reply structure that preserved argumentative context, and a hosting model that — while fragile — at least aspired to keep the URL alive. Discord and Slack fail on all four axes: not indexed by search engines, no stable URLs a non-member can use, flat channels that fragment context, and message retention policies that legally require deletion.
Compare the debugging economy this creates. In 2008, a senior developer hitting an obscure `iptables` conntrack bug would Google the exact error string and land on a LinuxQuestions.org thread from 2005 with the answer in reply #14. In 2026, the same developer hits the same class of bug, asks in a company Slack, gets an answer from a colleague, and that answer evaporates in 90 days per the retention policy. The knowledge exists for exactly one debugging session and is then erased.
This is the part senior developers should care about, because it explains why LLMs are simultaneously so good at old problems and so bad at new ones. GPT-4 and Claude are effectively trained on the corpus of the crappy-forum era — Stack Overflow, phpBB scrapes, LKML archives, mailing-list dumps. That corpus stopped growing around 2015. The last decade of engineering knowledge — the Kubernetes debugging tricks, the Rust async footguns, the specific ways Terraform state files corrupt — mostly lives in Slack, Discord, private GitHub Discussions, and screenshots pasted into Notion pages. Model quality on 2010 problems is superhuman; model quality on 2023 problems is noticeably worse, and this is downstream of where the discussion happened.
The HN thread's most-upvoted counterpoint is worth engaging with honestly: forums were also a moderation nightmare, a spam magnet, and a legal liability. That's true. The comparison isn't "forums vs. Discord in a vacuum" — it's "forums vs. the specific outcome we got, which is knowledge disappearing into private silos." Discourse (the platform, not the concept) is the current attempt to split the difference: modern UX, open-web indexability, but it requires an operator willing to run infra. Most teams pick Slack instead because it's zero-ops.
Three concrete implications, in decreasing order of how much your team can actually control:
Audit your own knowledge decay. If your team runs on Slack, calculate how much institutional knowledge you're losing on your current retention policy. A 90-day retention on a channel with 40 engineers is roughly 10,000 messages of context erased per quarter. The fix isn't extending retention — Slack search is bad enough that it barely helps. The fix is a lightweight write-up culture where the *answer* (not the discussion) gets copied into a searchable, indexable wiki within the same working session. GitLab and Basecamp have been public about doing exactly this for a decade.
Reconsider public forums for your own project. If you maintain an open-source library, the reflexive answer in 2026 is "start a Discord." Discord is a knowledge tarpit. Every question answered in your Discord is a question that will be asked again next month, because the answer is not on Google. GitHub Discussions is the low-effort compromise: indexed, threaded, stable URLs, and free. Discourse if you have the budget. The maintainers of Rust, Elixir, and Julia figured this out years ago; the JavaScript ecosystem is still learning it the hard way.
For LLM-assisted development specifically, notice that the tools you rely on are trained on a corpus that mostly predates the knowledge-in-Discord era. If Claude or Copilot is unusually bad at a specific problem, it's often because the answer lives in a private Discord you're not indexed against. This is a solvable problem — some teams are starting to ingest their own Slack history into RAG systems — but the industry-wide version of that fix is unlikely to arrive.
Smith's essay reads as nostalgic, but the useful reframe is that the forum era solved a specific problem — durable, searchable, community-maintained technical knowledge — and the industry has spent 15 years replacing it with tools that are better at engagement metrics and worse at every other thing forums were for. The forums aren't coming back. But the properties they had are worth naming explicitly, because every knowledge-management decision your team makes in 2026 is a bet on which of those properties you're willing to give up. Most teams give up all four without noticing.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.