Don't Take the Black Pill

A conference talk by Zig's creator that spends almost none of its runtime on languages or compilers. Kelley's subject is why software is worse than it was, and his answer is not technical: the field knows how to build robust software and has since before he was born. What changed is who gets to decide. The talk's second half is about what to do with that conclusion without curdling into the nihilism it names in the title.

The argument

The black pill is borrowed from incel forums, and Kelley walks through its mechanism carefully because the mechanism is the point. A community forms around a real grievance. The people who work through the problem leave, because once you've solved it you have no reason to keep commiserating. Newcomers therefore meet only the people who didn't. Over time the accepted wisdom becomes "give up, the game is rigged" — not because anyone argued for it, but because selection removed everyone who would have argued otherwise (01:07, 02:15).

He reaches for cory-doctorow's 2019 story Radicalized for the same shape in a different setting — a forum of people wrecked by health insurers, where "the people who got over their furious grief left the forum, chased away by its rage culture" (04:24). Five years after publication, the CEO of UnitedHealthcare was shot dead.

Quality was never sustained by craft alone. The central claim comes at 11:15: software quality "was maintained only through the will of software engineers combined with their accidentally strong negotiation position in the job market." Scarcity of programmers, not professional standards, is what let engineers win the argument with management. As the labor market normalizes, that leverage goes, and with it the ability to make things work.

The setup for that claim is a show of hands — who tried to make software better and got overruled because the business had other priorities. Over half the room (06:42). The coping strategies he then catalogues are all forms of what he calls benevolent non-compliance: forgiveness over permission, outright secrecy, misdirection, and the duck — deliberately planting an obvious flaw for management to find and remove so the real work goes unexamined (09:01). Every one is a gamble against retaliation, and which way the odds run depends on the job market.

Coercionware. The middle section builds from the transistor as "a mechanism by which something weak controls something strong" to the question of who decides what goes into the base terminal (12:22). Software that serves whoever is directly using it is the norm he wants; a browser is called a user agent for a reason, which is why ad blocking is legitimate and DRM is not. Software built to manipulate its users rather than serve them he calls coercionware, and the list runs through Mechanical Turk, bossware, advertising, push notifications, A/B testing, the YouTube algorithm, AI agents, and dark patterns (15:53–19:14).

The A/B testing objection is the sharpest and the least conventional: it "prioritizes outcome over empathy," and by replacing a design judgement with a metric it supplies plausible deniability for dark patterns. If the red button converts better because it frightens people, nobody wrote that down (16:58).

On AI agents: "AI agents promise to serve the user, but in reality they serve the AI company who controls the model and the closed-source software surrounding it. If you don't control the source code, you are not in control" (18:07).

Doomscrolling as an exploit. He refuses the framing of infinite scroll as a bad habit. Facebook's client is "categorically, fundamentally coercive… like an arbitrary code execution exploit running on our brains. Scrolling down is the program counter and the content that pops up is the instructions" (20:21). The consequence he draws is that a channel built to be exploited by its owner is also exploitable by third parties, and the 2017 ODNI assessment of the Russian influence campaign is what a hostile party using it looks like. The vulnerability has not been patched.

Black-pilled takes, named

Kelley makes a point of framing this as empathy rather than blame, then lists specimens including his own (22:37):

• "Computers were a mistake," said by career programmers. It denies all agency, and they are the ones making computers do things.
• Using AI you dislike because you believe you must. He describes a friend's team working 50-hour weeks with AI to eliminate other teams' jobs before it happens to them.
• Anubis — not the mascot complaints, which he dismisses, but the design. Making a cheap anonymous act deliberately expensive "whispers bitterly in your ear, 'There is no winning. There is only deferred loss.'" (24:58)
• His own "Site is trash now" comment about Hacker News, which he screenshots and admits is melodramatic.
• Withdrawing from free software because LLM companies scrape it. This one he argues against on the merits: the point of the commons is that high-quality free software constrains how coercive proprietary software can afford to be. Refusing to contribute surrenders that leverage and does nothing to the scrapers (26:03).
• "If I don't do it, someone else will," which he treats as the one that does the most damage to a moral compass.

The free-software argument is the useful one, and it sits directly against the position catalogued in llm-enshittification — same premises about scraping and labor displacement, opposite conclusion about whether to keep publishing.

The turn

Technology accelerates humanity along whatever path it is already on, so "if humanity is good, then technology is good" — and believing that is a premise you adopt, not a result you derive (28:20). His example is Stanislav Petrov, who in 1983 judged the odds of a false alarm at 50/50 and disobeyed anyway. At even odds the evidence decides nothing; something else did (29:27).

The closing section is about generations. Young people lack the coping mechanisms that keep older people from being hurt the same way twice, which is exactly why they are still willing to attempt what has already failed. "Our task as the older generation is to shield the young from those who would exploit their naivete without infecting them with our own accumulated pain and cynicism" (33:47).

He ends on the group of friends he lost touch with over Discord, and the fact that they are building a replacement rather than complaining — pointedly not naming it. "You only need a computer, a sense of curiosity, and time" (34:50).

Relation to other pages

The market-power thesis is a different diagnosis from enshittification. Doctorow's decay runs platform → business customers → nobody, restrained by competition, regulation, worker power, and antitrust; Kelley collapses that to worker power specifically, and to programmers rather than workers generally. Both land on the same object — software that got worse on purpose — from opposite ends.

Kelley's other public positions are on zig, including the anti-LLM contribution policy, which reads differently next to this talk: the same conviction that whoever controls the source controls the outcome.

Every quote on this page comes from YouTube's automatic captions in kelley-dont-take-the-black-pill and has not been checked against the audio. The [ __ ] redactions and >> [laughter] >> markers there are YouTube's, not the speaker's. Verify before quoting any of this anywhere it matters.