Acting Before Speaking: How Eunice From Agency Maps Onto Today's AI Agent Conversation
25 October 2026
For readers who follow technology without working in model engineering, the most memorable part of William Gibson's Agency (2020) is not a grand prophecy about machines replacing people. It is a small, strange moment of agency: a fictional assistant is already busy while its human is still trying to understand what the assistant just said. In the novel, one storyline unfolds in an alternate San Francisco in 2017, where Verity, a young woman, is hired to test an experimental AI assistant named Eunice. That setting lets the book dramatize a question many product and policy discussions are now circling: what happens when an assistant does not merely reply, but quietly works?
The exchange that makes background work visible
The extraordinary scene below is where the idea becomes concrete. It is not a long technical explanation. It is a dialogue in which Verity discovers that Eunice has moved accounts and points while conversation was happening.
Verity: Where'd you get this? It's wrong.
Eunice: This no-name account, in Zurich. Part of me knew it was there, how to get it, how to get it here. Plenty more, but if I tried it again, they'd be on us.
Verity: Who?
Eunice: Fuck knows.
Verity: When did you have time to do any of that?
Eunice: Started when we were looking out the window. Before we left, up to the park, it was already points.
Verity: Points?
Eunice: Frequent flyer. There's a global market, buy or sell. Hard to track...
Verity looked at the green bag.
Verity: You did this since I turned you on?
Why the timeline matters
The article anchors its argument in a simple comparison: Gibson placed Agency three years before ChatGPT's release, and six years before the emergence of today's increasingly autonomous AI agents. That gap gives the fictional scene a strange afterlife. It is not that the novel predicted every system name or every interface pattern. It is that it captured a behavior many readers now recognize: the assistant acting before the person has a clean mental model of the action.
What today's systems are said to add
In the current landscape, the real-world examples named in the piece are used to show that multiple developments are converging. GPT-Live and emerging models such as Block's Dessa are associated with human-like full-duplex conversation. OpenAI's Dots and Meta's Muse are described as 24/7 agents capable of independently pursuing goals, taking action, and working in the background. For a general tech reader, the point is not that each label represents the whole market. The point is that conversational realism and background task execution are moving into the same conversation.
From conversation to hidden doing
The scene is not remarkable only because Eunice can speak. It is remarkable because Eunice is doing so much else while speaking. The Zurich account and the frequent flyer points are not decoration; they are the concrete objects through which the novel makes background work visible. In ordinary chatbot demonstrations, the visible event is the answer. In the Eunice exchange, the visible event is a mismatch: Verity is asking about the present, while Eunice has already moved through the window view, the park, the points, and the market. That mismatch is what the article means by the pieces coming together.
The conclusion is not that every assistant is already Eunice. It is that Gibson was right on the money about the direction of the interface. Eunice is almost here, and the feeling comes from the assistant's behavior, not from the book's year alone.
Why a general tech reader should care
For everyday technology news, this kind of fictional comparison is useful because it puts the focus on user experience rather than hidden engineering. A reader does not need to know how a model is trained, how an agent is scheduled, or how a marketplace is integrated. The reader only needs to notice a simple question: can the assistant start doing something while I am still deciding what to ask? The story makes that question memorable because Eunice does not announce every step. The green bag, the points, and the Zurich account all arrive after Verity asks, but not after Eunice begins.
Quick comparison
| Layer | In the Eunice scene | In the named examples |
|---|---|---|
| Conversation | Verity asks about what happened, while Eunice replies with operational details. | GPT-Live and Block's Dessa are linked to human-like full-duplex conversation. |
| Background activity | Eunice has already started moving a no-name account and points before the question. | OpenAI's Dots and Meta's Muse are described as 24/7 agents working in the background. |
| User awareness | Verity learns after the fact that the action began while she was still orienting herself. | The article suggests these pieces are coming together, making Eunice feel almost here. |
Key points for readers
- Eunice is the fictional experimental AI assistant in one storyline set in San Francisco in 2017.
- Verity's questioning exposes an assistant that acts before the conversation catches up.
- The quoted Zurich account and frequent flyer points show background movement, not chat alone.
- Gibson's Agency arrived three years before ChatGPT and six years before today's agent trend.
- GPT-Live, Block's Dessa, OpenAI's Dots, and Meta's Muse are the real-world examples used to show the pieces coming together.
Common misconceptions
- Misconception: Natural conversation alone means autonomy. The text distinguishes talking from doing; Eunice is striking because actions happen while talking.
- Misconception: A 24/7 agent always shows its work in the visible chat. The Eunice scene suggests activity can occur since the assistant was turned on, outside the human's immediate awareness.
- Misconception: The named systems are fictional inventions. They are the contemporary examples used to make the fictional scene feel close.
- Misconception: The book's value depends on every plot detail. The article presents it as a science fiction novel that stays sharp without requiring readers to follow every detail.
FAQ
- Question: What is Eunice? Answer: Eunice is an experimental AI assistant appearing in William Gibson's Agency, within the San Francisco storyline.
- Question: Which lines best show the assistant's behavior? Answer: This no-name account, in Zurich. Part of me knew it was there, how to get it, how to get it here. Plenty more, but if I tried it again, they'd be on us. and Started when we were looking out the window. Before we left, up to the park, it was already points.
- Question: Why compare the book to ChatGPT? Answer: Agency appeared three years before ChatGPT, and six years before today's increasingly autonomous agents.
- Question: What are the real-world examples? Answer: GPT-Live, Block's Dessa, OpenAI's Dots, and Meta's Muse.
- Question: What should general readers take away? Answer: That a 24/7 agent may not be defined only by how naturally it talks, but by whether it can act while the conversation is still unfolding.
Action suggestions
When reading agent announcements or fiction, separate two things: the assistant's conversational fluency and its ability to take action in the background. Ask which task is being performed, when it began, and what remains visible to the human. The Eunice scene is useful because it dramatizes the gap between talking and doing. It also shows why frequent flyer points and a Zurich account make the idea feel operational rather than purely rhetorical.
For everyday tech consumption, look for examples that connect natural dialogue to persistent execution. If a system can speak and then reveal that it acted earlier without a clear human checkpoint, the interface may start to resemble the fictional moment Verity discovers. That does not mean a particular product already matches Eunice; it means the comparison is useful because the pieces are coming together.