Pris is the one who says it. Not Roy, not Batty with his tears in rain — Pris, the Nexus-6 replicant built for what the film calls combat and pleasure, cornered in a house full of chess pieces and taxidermy, delivers the line the whole genre has quietly organized itself around ever since: "I think, therefore I am." Descartes, borrowed by a machine, deployed as a legal argument. It's one of the best lines in Blade Runner because it asks the only question 1982 thought was worth asking about thinking machines: do they have the standing to ask for anything at all.
Earlier in the film, Captain Bryant has already answered a different question, and answered it so casually it barely registers as an answer. "Don't be an asshole, Deckard," he says, "I've got four skin-jobs walking the streets." Four minds, if Pris is right about what she is, reduced to a maintenance category. Not villains. Not even entities. A backlog.
We remember Pris's line. Almost nobody remembers Bryant's. Fiction gave us an extraordinary, unforgettable rehearsal of the wrong debate, and a throwaway, forgettable rehearsal of the right one. For forty years we've argued about whether machines can think, when the only question that ever mattered is what happens to the humans once the machines can do their jobs — a question Blade Runner answered in one dismissive sentence from a police captain and then never returned to, because it was never the interesting scene.
We're living the answer now. Not in an off-world colony. In pull requests.
Boris Cherny, the engineer Anthropic put in charge of Claude Code, its AI coding tool, says, without any particular drama, that he hasn't written code by hand since November 2025. He ships more than twenty pull requests a day — not twenty tickets closed, twenty separate units of software, reviewed and merged, mostly orchestrated rather than typed. Ask what this means for the discipline of programming and he reaches, unprompted, for the printing press: yes, a few scribes will always love the pen, but nobody sensible mourns the illuminated manuscript once everyone can read. Democratization. A generalist future. Coding as a background process, invisible, available to anyone who can articulate intent.
Jenny Wen leads design for Claude at Anthropic, after running design teams at Figma, Dropbox, Square, and Shopify. She says the discovery-mock-iterate process her discipline has been built on for thirty years is already dead, that engineers — not designers — are the ones forcing the change, and the question she's sitting with in public is whether AI will eventually surpass humans in taste and judgment.
Notice what both of them reached for. Not "what happens to the people who built their identity around this skill." Taste. Judgment. Democratization. Generalism. The philosophical question, in fresh vocabulary, arriving right on schedule — because it's, again, the more interesting scene. Cherny and Wen aren't being evasive. They're doing exactly what Blade Runner trained a whole culture to do: reach instinctively for the question about the machine's inner life, and let the question about the human's disposability pass by in a subordinate clause, the way Bryant let it pass by in his.
None of this is cynicism. Cherny and Wen, and the cohort they speak for, mostly believe this is good — and they're not wrong that something is gained alongside what's lost. They can see the impact more clearly than almost anyone, which is exactly why their peace with it lands so easily: society will figure it out. It's a genuine belief, not a dodge, and that's what makes it so hard to argue with. But the people equipped to catch Cherny mid-decoy, or to notice what Wen's answer sidesteps, are not a cross-section of that society. They're the same narrow population already living years ahead of everyone else, for whom a compressed timeline is simply Tuesday. Society will figure it out is true only if you ignore who is actually being asked to do the figuring, and at what speed.
Most people's experience of "AI" is a vacuum cleaner that no longer gets stuck under the sofa, a car that changes lanes by itself on a motorway trip taken twice a year, or a chat bot that answers questions. Judged at that pace, this looks exactly like every other adoption curve of the last century — television, the microwave, the smartphone — a slow, forgivable crawl from novelty to background furniture. Humans are built for that curve. Ten thousand years of practice absorbing change slowly enough that our mental models could keep pace with the world they described.
That's the trap. The domestic pace and the industry pace are not the same clock, and almost nobody outside the technology industry has been told this. There's a version of this mismatch that shows up every generation — an established cohort applying a working mental model to a world that's already changed underneath it, not out of stupidity but because the model used to be correct and nothing has yet forced anyone to notice it no longer is. Call it the old error: the lag between a new reality arriving and the cognitive frameworks that would let someone recognize it. Historically that lag was an inconvenience. You misjudged a market, misread a generation, wrote a think-piece that aged badly. The lag itself never used to be dangerous, because the underlying pace of change was slow enough to absorb an error in judgment before it compounded into anything irreversible.
The pace is no longer slow. Everyone actually building this — the Chernys, the Wens, the twenty-pull-requests-a-day cohort — will tell you plainly that the wave is close and it's fast. The rest of the world is still budgeting for the old tempo, still expecting the multi-decade runway every previous technology extended as a courtesy. This one may not extend it. A lag that used to cost you a bad opinion now costs you the chance to have had one at all.
A change nobody notices is, by definition, a change nobody has to answer for. That's the quieter cost of the frog in the water — not that it boils before anyone jumps out, but that no one is ever in the room where the temperature gets set. Bryant didn't convene a hearing before deciding four skin-jobs counted as a maintenance problem rather than four minds; nobody remembers a meeting, because there wasn't one. The decision simply arrived, already made, dressed as procedure. That's the shape of what's happening now — not a decision anyone took, but one that keeps accumulating in the space where a decision should have been.
Every technology that has previously reordered who is necessary and who isn't left some kind of trail behind it. Nuclear power got a treaty. IVF got a court case and a papal statement and a decade of hearings before the first clinic opened its doors. Even the factory floor, brutal as its transition was, got a labor movement — an adversary, at least, which is another way of saying it got acknowledged as a fight. You can find the minutes. You can find who said yes, and who said no and was overruled, and the date it happened.
Find the minutes for this one. Find the meeting where it was decided that a compressed timeline a narrow slice of engineers experiences as ordinary should also apply, without notice or consent, to a claims adjuster, a junior copywriter, a paralegal doing first-pass discovery. It doesn't exist, because the decision was never made in the singular. It's being made in aggregate, by ten thousand separate product choices, each individually reasonable, each shipped by someone simply doing their job well, none of them the moment anyone could point to afterward and say: there, that's where we agreed to this.
Governance exists almost entirely to audit decisions after the fact. It assumes a decision-maker, a moment, a document. It has no real answer for a transformation with no author — for agency handed over not by choice but by accretion, the sum of everyone doing their job competently and nobody responsible for the total.
Be careful what you wished for is usually a warning about getting exactly what you asked for and hating it. That's not quite what happened here. We didn't get what we wished for. Fiction wished, on our behalf, for a machine sophisticated enough to ask whether it deserved rights — a thrilling, flattering problem, because it made us the judges. What arrived instead was a machine sophisticated enough that the question of rights never came up, because it didn't need to think of itself as a person in order to do the job a person used to hold. We spent forty years rehearsing the trial of Pris and missed that Bryant's sentence was the one that would actually happen to us.
There won't be a Voight-Kampff test for whoever comes for the next round of jobs. There won't be a scene where someone is asked to prove their inner life matters. It will simply be a background process, invisible, running while everyone who was supposed to notice was still watching the wrong movie.
The frontier isn't wrong to think something better might be waiting on the other side of this. But they're moving at machine speed, and the rest of us are still fog — slow, human-paced, unable to see more than a few feet past the actual shape of what's coming. Between those two speeds there's a gap, and gaps like this don't hold still forever. They close, or they tear.
So here's the only question worth carrying out of this: if no one convened the meeting where this was decided, is there still a meeting we could convene now — and if there isn't, what does it mean that the only people who could call one are the same people who benefit most from it never happening?