Bedor argues that asking another human to spend their finite attention on your PR, bug report, cover letter, or DM carries an implicit obligation: you must have spent some of your own attention first. AI-generated submissions collapse production cost to near zero while leaving consumption cost untouched, violating the reciprocity that made the contract work.
By submitting the essay and driving it to 1,535 points, the HN community amplified the thesis that demonstrated effort is a moral prerequisite for requesting attention. The high score and 467 comments signal broad agreement that the norm has been violated at scale.
The editorial reframes Bedor's intuition through Amotz Zahavi's 1975 handicap principle: effort signals stay honest only when they are expensive to fake and more expensive for low-quality signalers than high-quality ones. LLMs have collapsed the production cost for low-quality signalers specifically, which is precisely the failure mode the handicap principle predicts will destroy the signal.
Top comments converge on the same pattern across domains: curl-style CVE reports citing nonexistent functions, take-home coding submissions the candidate clearly never read, and college essays that can no longer be assigned. The consistency across unrelated roles is treated as evidence that the broken contract is systemic, not anecdotal.
Tom Bedor's essay "If you are asking for human attention, demonstrate human effort" hit 1,535 points on Hacker News, with the comment section reading like a maintainer support group. The piece is short and the thesis is one sentence: when you ask a human to spend their finite, non-fungible attention on your bug report, your pull request, your cover letter, or your DM, you owe them evidence that you spent some of yours first. Bedor is reacting to a year of AI-generated PRs with hallucinated APIs, bug reports that don't reproduce, cold-outreach emails written by ChatGPT, and grant applications that all sound the same because they came from the same temperature-0.7 sampler.
The top comments tell on the industry. A curl maintainer-style anecdote about CVE reports generated by AI tools that cite functions that don't exist. A hiring manager describing a take-home submission where the candidate clearly hadn't read the code they submitted. A professor who can no longer assign essays. The pattern is consistent enough that the comments converge on the same intuition: the implicit contract — *I burned an hour writing this, please burn ten minutes reading it* — is broken. Production cost on one side has collapsed. Consumption cost on the other side has not.
Bedor doesn't reach for the biology, but the biology is what makes the essay's claim provable rather than just resonant. The reason "demonstrate effort" worked for the last few thousand years of human communication is that it was a *costly signal* in the formal sense — Amotz Zahavi's 1975 handicap principle.
Zahavi's principle, now bedrock in evolutionary biology and signaling economics, says that for a signal to remain honest in a system where deception is possible, the signal must be expensive to produce — and *more* expensive for low-quality signalers than for high-quality ones. The peacock's tail is the canonical example: it's metabolically ruinous and makes you slower to a fox, which is exactly why peahens trust it. A weak peacock can't grow one. A stotting gazelle who literally jumps in the air in front of a cheetah is broadcasting "I am so fit I can afford to waste energy taunting you" — and cheetahs respect the signal because faking it gets you eaten.
The entire pre-2023 etiquette of human attention ran on the same logic. A well-written cover letter signaled effort because it took hours. A thoughtful bug report signaled effort because reproducing the bug was hard. A long, technically detailed email signaled effort because writing technically detailed prose is hard. Reviewers, maintainers, hiring managers, and editors weren't really evaluating the artifact — they were evaluating the cost of production as a proxy for the sender's seriousness. That proxy worked for the same reason peacock tails work: forgery was metabolically expensive.
LLMs are a forgery technology that drove the metabolic cost of a 2,000-word technical document to roughly the price of a Diet Coke. The honest signal didn't get weaker; the dishonest signal got cheap enough to mass-produce. Biology has a name for what happens next: signal erosion. When forgery costs collapse, receivers either (a) stop trusting the signal entirely, (b) raise the cost of the signal until honesty is restored, or (c) switch to a different signal that's still expensive to fake. We are, collectively, doing all three at once and badly.
The HN thread is full of practitioners stumbling into option (a). Maintainers closing AI-flavored PRs unread. Editors rejecting any submission with em-dashes. Reviewers who admit they now skim the first paragraph for tells and bin anything that smells like sampled text. This is the bird that stopped trusting tail length — efficient in the short run, brutal for any legitimate signaler who happens to write like a transformer. The collateral damage is non-native English speakers, whose careful prose has always sounded slightly LLM-shaped, and who now get filtered out of the front of the funnel.
Option (b) — raising the cost — is what Bedor is actually proposing, and what most thoughtful maintainers are quietly doing. "Show me the reproduction" is a costly signal: an LLM can guess at a repro but can't actually run your code. "Get on a call" is a costly signal: synchronous time is the one resource that hasn't deflated. "Show me the commit history of you wrestling with this" is a costly signal: the artifact of struggle is hard to fake even if the final code isn't. The maintainers who are surviving the slop wave aren't reading more carefully — they're demanding signals that LLMs structurally cannot produce cheaply.
Option (c) — switching signals entirely — is the most interesting and the most underdeveloped. In biology, when one signal erodes, populations often migrate to a *combination* signal: not just the tail, but the tail plus the dance plus the lek location. The human-attention equivalent is reputation graphs, real-name introductions, and embedded co-presence. Discord servers vouching for contributors. GitHub histories with five years of unrelated work. The return of the warm intro. None of these are LLM-proof in principle, but they're orders of magnitude more expensive to fake than a single document.
If you run a code review process, an OSS project, or a hiring pipeline, you are now a receiver in a degraded signaling environment, and your filters need redesign. Three concrete moves, in order of how cheaply you can ship them:
First, stop evaluating artifacts and start evaluating processes. A PR description that explains the three approaches the author considered and why they rejected two is a costly signal — LLMs can fabricate it, but the fabrication breaks under one follow-up question. Add a CODEOWNERS-level requirement that PRs include a "what I tried that didn't work" section, and treat its absence as the new failing-CI. The same logic applies to bug reports ("steps I took to isolate") and to take-home interviews ("git log shows the path you took").
Second, reintroduce synchronous gates at the points where stakes are highest. The right place isn't the top of the funnel — keep that wide and welcoming — it's the point at which you're about to spend real reviewer time or extend an offer. A 20-minute live walkthrough of a take-home submission catches more LLM-laundering than any plagiarism detector. A pair-programming session on a real bug catches more than a one-way video. The cost is paid by *you* as much as the candidate; that symmetry is the feature, not the bug.
Third, build reputation infrastructure that compounds. Internal-only kudos systems, public commit histories with non-trivial tenure, and contributor graphs that surface "this person has been around" are exactly the multi-signal stacks that biology evolved to defeat single-signal forgery. The teams that will run hot in 2027 are the ones whose internal social graph is dense enough that a new contributor's first PR arrives pre-vouched.
The deeper bet inside Bedor's essay is that human attention is about to become the scarcest resource in software — scarcer than compute, scarcer than capital, scarcer even than senior engineers. LLMs didn't create that scarcity; they revealed it by removing the friction that used to ration access to it. Every system we built that depended on the cost of writing being a filter — peer review, cold outreach, hiring, code review, grant applications, journalism pitches — is now an open-loop amplifier with no governor. The next five years of platform design is going to be a slow, expensive search for new costly signals. Biology has been running this search for half a billion years and the answer it keeps converging on is the same one Bedor lands on: if you want my attention, spend some of yours. The medium of that spending is what's up for grabs.
Top 10 dev stories every morning at 8am UTC. AI-curated. Retro terminal HTML email.