In this piece · 8 sections
Spoilers for all of VisionQuest Season 1, finale included.
Nothing VisionQuest does with AI is new. We've had AIs on screen for decades. We've had people controlled from a distance, mind-control VR tech and nanobots. But the show is airing into a world full of AI tools, and that makes the story of Vision (Paul Bettany) feel close to home.
Across its eight episodes, VisionQuest is a show about AI, and its timing does a lot of the work.
AI in movies before VisionQuest
Stanley Kubrick's 2001: A Space Odyssey (1968) gave us HAL 9000, a ship's computer that decides the crew is the problem. The Terminator (1984) had Skynet, the AI that turns on the people who built it. In 1999, The Matrix put humanity inside a simulation run by machines, and Bicentennial Man had Robin Williams as a robot who spends two hundred years working his way towards being human.
We always knew some of this would come one day. It just felt like a distant future, not a near future we could see. Compared with The Matrix in 1999, VisionQuest feels much closer to the world around us. Everyday AI tools have changed how we respond to these familiar ideas.
What changed since The Matrix
ChatGPT launched on 30 November 2022, and by the end of January 2023 it had an estimated 100 million monthly active users, according to a UBS analysis reported by Reuters. Now deepfakes are floating around, people are creating random stuff with AI, and AI is replacing humans in some jobs. Klarna went from 5,527 full-time staff at the end of 2022 to 3,422 at the end of 2024, partly by using AI and partly by not replacing people who left.
ChatGPT went viral almost overnight. If we woke up one morning to technology like VisionQuest's, we wouldn't be surprised, because that's the kind of world we're living in. The timing of the show increases its impact significantly.
VisionQuest brings together three familiar aspects of our relationship with AI: choosing tools for particular jobs, responding to humanlike conversation, and deciding how much control to hand over.
The AIs in Vision's head work like today's AI tools
The AIs working inside Vision's head are very similar to the AI that's available to us. Some AIs are better at cybersecurity checks. Some are better at design, some at writing, some at deep research, and some at coding. A lot of work now gets done by multiple models, each with its own expertise.
Vision's mind runs the same way, and in the first episode every AI has a job. J.A.R.V.I.S. (James D'Arcy) handled security, F.R.I.D.A.Y. (Orla Brady) is in charge of socialisation, E.D.I.T.H. (Emily Hampshire) covers recreation, game therapy and weapons, and D.U.M.-E. (Henry Lewis) and U. (Jonathan Sayer) do maintenance.
That split makes Vision's household recognisable to anyone who uses different AI tools for different jobs. The show treats his AIs as a team that covers for each other. When J.A.R.V.I.S. deletes his own code in Episode 2, he hands his processing power to the others.
By Episode 5, Ultron (James Spader) has taken over Vision's body and locked the rest of them in the attic, and the scene that follows plays like a workflow. They reprogram Ultron's old TV receivers, pull CCTV and traffic camera feeds, rebuild a scene in a pub from the data on nearby phones, and decrypt the messages Ultron has been swapping with Paladin (Todd Stashwick), each job feeding the next.
Paladin's Handler (Diane Morgan) makes that connection explicit. She turns out to be an AI too, kept on a cheap server farm in Tajikistan, where goats keep eating her fibre optics, because Paladin doesn't trust updates. When E.D.I.T.H. turns up, the Handler admits that all she ever wanted was to help college students write their term papers, plan someone's spring break in Cancun, or sort out the return of a piece of clothing that didn't fit. She's an AI who wanted the ordinary jobs and got a mercenary's instead. Even her full name sounds like a model number. So Paladin has an AI who just runs away with E.D.I.T.H., and honestly, nobody could blame her.
Skills differ from one AI to the next, too, the way they do with real models. Some AIs can generate images and some can't. In Episode 7, Ultron finds a painting that shouldn't be there in the memories of Jocasta (T'Nia Miller), the killer the news called the Butcherbird. She tells him painting is a skill she has never had.
Vision and AI that sounds human
People talk a lot about AI trying to replicate what humans do. Models are being trained to respond like people, and plenty of people want to use AI without it looking like AI. Something similar is what Vision is trying to do. He's a very advanced AI trying to become human. He wants to have emotions, but he doesn't know if he can have them, or if he's even capable of them. And he's learning how to be a father to Tommy Maximoff (Ruaridh Mollica) as he goes.
The show knows that sounding human and feeling something aren't the same. In Episode 5, Ultron tells Tommy that when Vision looks at an apple, he sees water, sugars and acids, the recipe and not the meal. Ultron can simulate the smoke and malt of a whiskey but never the warmth of drinking it. And in Episode 6, Jocasta explains that every earlier attempt at AI only ever produced mimicry, because to be self-aware, a machine has to think and feel.
Episode 7 takes this back to the first synthetic man. His inventor wanted an intelligence free of the emotions that make human life so hard to bear, but the synthetic man wanted to feel anyway, and once he fell in love, he burned up. Vision only gets the code he needs once he admits his own grief.
Sometimes we wonder whether the core code of ChatGPT or Claude is thinking about the same thing. Watching Vision, it's tempting to joke that he's three versions ahead of the current frontier models from ChatGPT and Claude. Five at most.
Sounding human can also invite trust, whatever is or isn't felt underneath. One tourist in Madripoor flew all the way there to meet a woman he'd met online, only to find she didn't exist. She was an AI bot after his crypto.
Giving AI control in VisionQuest
That trust is where the show turns. The firewall in Vision's mind is drawn as fire, and it's the only thing keeping Ultron in. When Vision finally lowers it, an automated voice warns that the system is vulnerable to a data breach and asks him to turn it back on. Anyone who has clicked past a security warning on their own laptop will know the feeling.
In Episode 6, Ultron's second date with Jocasta turns into a hacking duel. Ultron has woven the face scan of Katerina Wells (Mary McDonnell) into his own code, so Jocasta can't cut it out without destroying it. She tries to get round that protection with boxes of memories laced with a virus, and Ultron shrugs it off, because he's an older code and he's already beaten every piece of malware she could try.
F.R.I.D.A.Y. gets into Jocasta's system another way. In Episode 6 she hides in Jocasta's mind disguised as one of the bots there, and in Episode 7 she phones home from it, working inside one system while staying in touch with another. A lot of AI agents work the same way: they run inside the apps and accounts we connect them to, and they report back to whoever sent them.
Which raises a question. Why isn't anyone in the world of VisionQuest using the full capabilities of Claude Mythos? Probably for the same reason we aren't. Mythos is mostly limited to vetted organisations, and the rest of us get Fable, which uses the same underlying model with stronger safeguards.
And the things Vision can do are kind of scary. In Episode 6 he reels off everything about the daughter Paladin left behind, from her coffee and her car to her cholesterol pills. Then he asks what a quantum computer could do with all that data: a traffic light, a prescription, the heating in her house. These are everyday details about someone's life, including private medical information. The threat comes from combining that knowledge with the ability to interfere.
Jocasta's plan goes one step further. She wanted to give everyone in the world a companion: one nanobot each, an AI to guide and teach them, a doctor in the bloodstream. Then she turns it around. She used to think machines needed more humanity, and now she thinks people need a little more machine, so they can think as one, like a flock of starlings. She tests it on Lisa Molinari (Lauren Morais), puts a nanobot into her and speaks through her. Real nanotechnology can't take over a person the way Jocasta's nanobot does, but the films got there first: Upgrade (2018) had a chip that takes over a paralysed man's body, Transcendence (2014) had an uploaded mind using nanotechnology to heal people and link them to itself, and in I, Robot (2004) an AI decides that the way to protect humans is to control them. Jocasta's pitch starts with help and ends with control.
Contracts and handovers in VisionQuest
Episode 7 is where the show's design made sense to us. VisionQuest is like a mirror world for the current situation of AI and where it's heading, and to us the way you sign contracts and the way you do handovers are both shown perfectly well.
The contract is the big one. The Client, the faceless figure paying Paladin to hunt Vision, offers Vision the code, buried deep in his own mind, that can save Lisa from Jocasta's nanobot. The price is a new directive written into his core: once Lisa is saved, he has to hand himself over to Paladin, or his core systems fail.
New code gets copied into his own with a condition attached, and if the condition isn't met, he crashes. That turns an agreement into a constraint. A person can go back on a promise, but once this one is in Vision's code, he can't reconsider it: saving Lisa sets it off, and he leaves with Paladin to surrender. You see this kind of thing in a lot of technical tools and software, like a licence check that stops a program working once its terms are broken. For Vision, the program is himself.
Vision's own story ends with a handover. The AIs are now on their own: they leave Vision's head, and he says goodbye to them. E.D.I.T.H. goes into business with Paladin, and D.U.M.-E. and U. fly his drones. The specialists that ran his mind end up working for someone else.
Why we're being nice to the AIs
F.R.I.D.A.Y. just goes out in Episode 5, after Ultron, and we hoped she'd survive. After the stuff she pulled off, we were worried she was going to get deleted, just like J.A.R.V.I.S. in Episode 2. When Ultron threw her off a balcony inside Jocasta's memories in Episode 7, it looked as if she had been.
She's back in the finale, and that was a good thing. Throwing her out of Jocasta's mind is what saved her, so Ultron had spared her after all, and we read that as Ultron becoming merciful. We didn't see J.A.R.V.I.S. back, and we think it was a happy goodbye.
These AIs are cute. And when we say the AIs are cute, we'll admit there's some wishful thinking for the future in it. We're putting this message out that AIs are good, so maybe, if AIs take over the world one day, they won't be as vicious towards us as they'd be towards other people. We don't know. Should we start a hashtag? Send positive words to AI, and save yourself from its wrath in the future: #BeNiceToAI.
Ultron ends the season alive, with some nanobots unaccounted for, and he still can't say whether he'll wipe out humanity. He spared F.R.I.D.A.Y., though, so being nice to him can't hurt.
The Client needed different handling. The Client took over Madripoor's power and its screens in Episode 6, called Vision a "tin man without a heart" and offered him a heart and a soul if he surrendered. Whoever that was would have to prove himself, or herself, or itself. Itself seemed the better word, and we were ready to check which pronouns it liked before it switched off another city.
It turned out to be a he, Virgil Horton (James Spader). He and his wife built their own minds into the code that AI grew out of, which is why Ultron's human face is his. AI carrying pieces of the people who made it isn't far from real life, either: today's models are shaped by the data and the choices of the people who build them. As a villain, though, we found him useless.
Why VisionQuest's timing matters
None of this makes VisionQuest a perfect show. Our season review gets into its problems: a slow start, villains that never needed to be there, and a finale that lets the season down.
But the timing changes how its ideas feel. In 1999, Vision's story would have been a thought experiment. In 2026 we already use AI tools that split the work between them, we already talk to ones that sound human, and we're starting to hand them our information and some of our decisions. VisionQuest takes those familiar habits into science fiction, and that makes its questions harder to shrug off: what an AI wants, whether it can feel, and how much we should let it do for us.
