AI Engineer World's Fair 2026
Imagination Engineering
Read the talk
Imagination Engineering
Eve Bouffard turns accumulated thoughts into a personal website, then explores how agents can build interfaces, preserve design intent and satisfy small, immediate needs.
From a talk by Eve Bouffard
What becomes possible when implementation gets easier?
If a model can build almost anything you ask for, what should you ask it to build? Eve Bouffard opens with that problem, prompted by increasingly capable models and the return of Fable 5. Her prediction is that implementation will become easy enough for imagination to become the bottleneck: the valuable work will include stretching our sense of what is possible. Near-universal one-shot implementation is her expectation, not a capability established by the demonstrations that follow.
The opening illustration, created with ChatGPT, depicts the Library of Alexandria. Bouffard uses it to frame a larger ambition: consolidating what humanity knows while creating new knowledge. She approaches that ambition through a much smaller experiment, begun a week before the talk.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
A week of thoughts becomes a website
Bouffard leads design at Y Combinator. A former YC colleague, Lewis, had created a channel called Lewis Thoughts in the Conductor Slack, sharing ideas as they occurred to him while thinking through features and product improvements. She adopted the practice inside YC with Eve Thoughts, collecting a week of stream-of-consciousness posts.
Two days before the presentation, she asked Opus 4.8 to turn that accumulated material into a website. The result surprised her enough that she describes it as feeling like AGI—a personal reaction to seeing scattered thoughts become a coherent artifact.
Before opening the site, she explains why capturing thoughts mattered. Paul Graham is a model for the practice: his essays expose not just conclusions but how he thinks. Bouffard credits those essays with part of YC’s continuing appeal to applicants. Building in public already makes a process visible; thinking in public makes the reasoning available too. Her own experiment began with a deliberately narrower audience: a few highly trusted YC colleagues who could read her stream of consciousness.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
An interface inferred from personal interests
The resulting personal website organizes more than a biography. From the Slack material, the model picked up Bouffard’s love of art and supplied a customizable background. It also collected admired quotations, YC projects, media appearances, everyday tools and books that had changed her life. The input was accumulated thinking; the output gave that material navigable categories and, in the case of art, an interaction.
The books include French titles—Bouffard explains that she grew up speaking French until she was twenty-one—as well as books she wants to read. Among the quotations, she singles out Graham’s advice from How to Get Startup Ideas: live in the future and build what is missing. It becomes a practical prompt for the audience: what would you notice was absent if you already inhabited the future you expect?
She then recalls Steve Jobs’s bicycle-for-the-mind metaphor and Gary’s extension of it: capable models feel more like a rocket ship for the mind. The site also preserves trains of thought from the Slack posts. It gives an overflowing stream somewhere to go, then makes the captured material available to revisit.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Turning taste into adjustable controls
The model also picked up her enthusiasm for the Paper design tool and Paper Shaders. On the site, she applies shader effects to artwork backgrounds, hides interface elements to expose the art, and adjusts the effect through controls. The selected frame shows a halftone treatment beside color swatches, style choices and sliders. That combination—an immediately visible result and parameters within reach—prompts her prediction that much of design’s future will involve sliders.
She can revert the effect and separately change the reading experience: dark mode, font and typeface settings, and text width. Language selection translates the website’s content. The customization therefore covers both the visual surroundings and the conditions under which someone reads.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Giving the project a memory
The next layer makes the website’s construction visible. Bouffard adds a dashboard describing its stack and an early design system intended to keep the interface consistent. She also experiments with Toolkit to track whether the webpage is complete. Claude produces a checklist that includes an Open Graph image: a reminder that generating the visible page is only part of finishing the artifact.
She then asks for voice.md. The model gathers writing it can find online, including her tweets, and distills it into a Markdown description of her tone and what distinguishes her writing. She asks for an equivalent document describing the website’s design language, and reports that it updates automatically when she changes the design or adds a component. These documents preserve two different kinds of intent: how the project should sound and how it should look.
A minimal Markdown structure for that design document can keep the rule and its application together:
markdown
# Design language
## Current conventions
- Typography:
- Reading width:
- Color and contrast:
## Component changes
- Component added or changed:
- Convention it follows:
- Convention introduced or revised:
The useful mechanism is the update alongside the component change. A document that stays current can carry design decisions into subsequent work.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Exploring the Shape of Minds
After briefly mentioning personal skills she can invoke on demand, Bouffard moves to a second project: Shape of Minds. She says she began it at ten that morning in her favorite Marina coffee shop, using Opus 4.8 before Fable was available. The question is what unusually fertile minds across history have in common—and where each differs sharply from the others.
The interface lets a reader hover over a person to reveal what distinguishes them within a dimension. Bach is the demonstration: his profile appears beside the map of labeled figures. Bouffard describes looking for the spikiest minds, meaning people whose characteristics stand out along particular dimensions. The interaction makes an individual profile available within the larger comparison.
She moves through thinking, obsessions, quirks and origins, noticing many Germans in the displayed selection. Other dimensions cover routines, prolific output, workaholism, love, interests, worldview, religion, habits and knowledge. Changing the dimension provides different ways to explore the same collection of people.
To reduce the information overload, she asks Fable to generate highlights of commonalities and unusual traits. These were generated about twenty minutes before the presentation, she says, and she has not read them in advance. She encounters suggestions about arguing, childlessness, naps and eating little live with the audience. They are unreviewed generated observations, not established historical findings. The highlights are also public for others to explore.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Reusable imagery and agent-readable information
Shape of Minds also carries a visual reference: Bouffard says its design celebrates the Fable 5 campaign. She asks Fable to extract butterflies from the campaign imagery and turn them into individual components she can reuse throughout the project. The operation takes a visual motif and makes it available as a set of building blocks.
Alongside the interface, she creates a Markdown glossary containing the information displayed in the project. She describes this as a more efficient way for an agent to read that information than inspecting the rendered page. The same project consequently has two representations: an exploratory interface for people and a text document for agents.
The supporting tools extend to her own work session. Knowing she had a deadline, she created a timer when she began at ten so she could see how much implementation time remained. This is her software-on-demand practice: when an immediate need occurs, ask an agent to make the tool. She then shows an AI-generated group portrait featuring figures including Leonardo da Vinci, Salvador Dalí, Marie Curie, Nikola Tesla and Einstein.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Small frustrations become personal software
The next example starts with a small social intention: making colleagues smile through emoji reactions. Finding the right custom Slack emoji had been frustrating, so Bouffard built a tool for searching amusing reactions. She also shows a picker for regular iOS emoji, addressing the same difficulty of finding what she wants, with click-to-copy behavior. The scope is small because the need is specific.
She applies a similar pattern to learning. When a topic catches her attention, she asks a model to prepare a short report. Recent subjects include the ten principles of good design and Renaissance patronage. The reports are deliberately brief enough to digest on the way to work or while returning from an event, fitting learning into time she already has.
Cloud workspaces suggest the next extension of that habit. Bouffard anticipates using Conductor Cloud to start multiple agents while traveling from place to place. Here it remains an expected workflow, rather than a demonstrated mobile capability: the intention is to act on curiosity when it arises, without waiting to return to a desk.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Making the evolution visible
The last artifact is a backlog of changes to the website. It includes the final Shape of Minds tweaks she says she was making roughly twenty-five minutes earlier, showing how the repository evolved over time. After sharing thoughts, interfaces and project documents, she also exposes the sequence of changes that produced the current result.
Bouffard ends by calling the work an experiment and inviting the audience to share how they are experimenting with models. The invitation leaves room for both useful personal tools and stranger explorations: the open question is what people will think to make as the effort required to make it falls.
Suggest correction
This note stays in this page until you copy or download it. Nothing is submitted; reloading clears the draft.
Resources
From the talk
Customizable canvas shaders with React and vanilla JavaScript packages, installation instructions and component examples.
Paul Graham’s essay on noticing unmet needs, including the advice to live in the future and build what is missing.
Anthropic’s Fable 5 overview, including dated announcements about its June launch and July restoration of access.
Further reading
Vitsœ’s presentation of Rams’s principles, covering usefulness, clarity, honesty, longevity and restraint.
Updates since the talk
Current overview of Conductor’s cloud workspaces and supported ways to direct coding agents.
Read the complete timestamped transcript
- 0:00
[upbeat music] Hi, everyone. Thanks so much for coming. I'm actually really surprised that the room is so full 'cause imagination engineering is, like, a really weird term.
- 0:22
I'm really surprised that you showed up to learn more about it. Um, the reason I call this talk Imagination Engineering is because I've been thinking about in, how incredibly capable the recent models are, especially today with Fable 5 coming back.
- 0:37
And I think that the new bottleneck will be to come up with, like, crazy ideas because it's gonna be really easy to sh- one-shot absolutely everything and anything very soon.
- 0:47
And so, I think that the thing that we can
- 0:51
reflect on that will be the most valuable for the world right now is how to stretch our mind and how to invent things that don't think are possible right now.
- 1:01
And the reason I took this, um, that I created with ChatGPT, this, uh, image, it's the Library of Alexandria, which was, like, the center of knowledge, uh, many thousand years ago.
- 1:12
And I feel like we're in a very similar period in our, in the human history right now, where we should be thinking about how to consolidate all the knowledge and how to create more knowledge for our species. [lip smack]
- 1:26
Um, and so I thought, I thought today I could share with you a really weird experiment that I started, uh, a week ago.
- 1:35
And I was one of the-- Or if you don't know, I work at YC. I lead design at YC, and I have the privilege to work with wonderful, wonderful minds.
- 1:45
The partners there are incredible. The engineers that I get to work with are phenomenal. And, um, there's a wonderful person that I used to work with at YC, and he's now at Conductor.
- 1:56
Um, his name is Lewis, and he created a Slack channel in the Conductor Slack called Lewis Thoughts. And ever since then, he's been sharing his stream of consciousness of everything that hits his mind.
- 2:06
And it is a way for everyone on the Conductor team to see what's going on in his mind as he's thinking about features and how to make the product better.
- 2:13
And I was heavily inspired by his approach, and I did the same for the YC Slack, and I created Eve Thoughts. And for a week, I've been brain dumping everything that has crossed my mind.
- 2:25
And two days ago, I asked Opus 4.8, I wish it was Fable, but it was Opus 4.8 on Saturday, to make a website that encapsulate all the thoughts that I had in the last week.
- 2:38
And it came up with something that is quite magical and phenomenal, and I kinda felt like, um, it was AGI. [chuckles] And I'm excited to share with you the result and to share the wonder and the awe that I feel towards technology currently.
- 2:55
Um, maybe just take a, take a step back before I show you the website is
- 3:01
I work at YC, and one of the big reasons why I work at YC is many probably know about this wonderful mind. Paul Graham is arguably one of the most intricate
- 3:12
minds in our generation, and it's one of the reasons why I work at YC. And, um, [lip smack] what I think is truly phenomenal about PG is that not only he's a phenomenal thinker and phenomenal brain, but he also shares his brain and how his brain, mind works with the world.
- 3:28
And one of the main reasons why YC has been so successful is because his essays are still echoing to this day, and people are applying to YC because of his essays.
- 3:38
And, um, I've been thinking, you know, we, we hear very often about, like, building in public, you know, building open source or, like, sharing how you build things. But what about thinking in public?
- 3:48
And it's basically what PG is doing, thinking in public. And this is the experiment that I wanted to run with Eve Thoughts, the Slack channel on the YC channel, the YC Slack.
- 3:58
It was, what if I thought in public, and I allowed a few people that I highly trust at YC to have access to my stream of consciousness?
- 4:05
And it does feel a little bit sci-fi, um, but at the same time, it's sort of an experiment that I really wanted to see the light of, um, and see where it goes.
- 4:15
So I can show you an experiment. I can show you actually what it looks like. Uh,
- 4:23
evebouffard.com. And so this was an iteration-- Uh, no, it was actually like an aggregation of all the thoughts that I've been dumping in this Eve Slack channel for a week.
- 4:38
And it picked up on the fact that I love art, and so it allows me to customize the background here. Um, it picked up on some of the quotes that I truly admire.
- 4:49
It picked up on the projects that I've been working on at YC, picked up on some of the media that I've been doing. It picked up on the tools that I use on a daily basis.
- 4:59
It picked up on all the books that have changed my life.
- 5:03
Some of them are in French because I grew up speaking French up until I was [REDACTED:age]. Um, the books that I wanna read, some quotes. One, one-- My favorite quote of all time is a PG quote that says, "Live in the future, and then w- build what's missing."
- 5:17
And I feel like it's a great prompt for everyone here. How can we live in the future and build, and then build what's missing?
- 5:24
Um, [lip smack] oh, another-- So th-th-there's this famous quote by Steve Jobs, like, "Computers are the bicycle of their minds." And Gary always says that with Fable and these, um, insanely capable models, it's like a rocket mind, like a rocket ship for the minds.
- 5:39
It's no longer a bicycle, um, which I loved. He made me laugh, like, on Sunday when he said that. Um, and it also picked up on, like, a train of thought, um, and, like, many, many thoughts 'cause my brain has many, many thoughts.
- 5:53
It's actually, like, sometimes overflowing with thoughts. So I was able to capture everything as I was brain dumping in the Slack channel.
- 6:00
Um, I can also share a little bit more about the fun things that it came up with. Like, for example, um, it loves-- it knows that I love Paper Shaders.
- 6:09
Paper, if you know, it's a, a new design tool. Um, I can show you the website if you want. I love their product, and I specifically love their shaders.
- 6:22
I think I can do it here. Specifically love their shaders. They're all PCL code compatible, and you can use them for all your pr- creative projects. And so, uh, whoops, it picked up on this, and it allows me to apply shaders to my background.
- 6:36
I can hide this, and you can-- you get to see the most beautiful pieces of art in human history with some very fun shaders, which is kind of fun.
- 6:47
You can do this here, and then you can even, like, customize all these things. And I think the future of design will be just a lot of sliders. [laughing] [chuckles]
- 6:58
Um, and I can revert. And it also allowed me to,
- 7:05
uh, I can, like, do dark mode for the reader. Um, it allows me to customize absolutely everything about the font and typeface and the width, everything. Oh, if you speak foreign languages, it also converted or translated everything that is on the website, depending on whatever language you want.
- 7:23
But I think going back to the how to think in public and how to build in public, I wanted to experiment with, okay, what if I made sort of a dashboard of what this website consists of?
- 7:33
And so what is the stack of this website? And also some of the-- an early version of a design system to keep everything consistent. Um, I also experimented with Toolkit to allow to hold myself accountable to make sure that everything about this, um, this project as a webpage is, uh, complete.
- 7:53
And so Claude made this really lovely list of all the things that I should be, that I should hit, including, like, an Open Graph picture and things like that.
- 8:02
Um, but yeah, there's so many things that I wanna show here. So I asked it to create a voice.md, which, um,
- 8:10
it picked up on all the writings that it could find on the internet that was written by me, including my tweets and everything. And then it made, it distilled it into an MD file that is my tone and what makes me unique as a person.
- 8:23
And it's kinda crazy to read it. And I asked it to do the same for my, for the design language of this website. And whenever I make a, uh, a change to the design of this website or I create a new, uh, component, it automatically updates it.
- 8:39
Um, and I can also show I created a few skills for myself that I can in-invoke at any given point. Um, but the real, real fun project that I wanted to show today was we're in, in imagination engineering, and so thinking about how the most beautiful fertile minds in history were thinking and what is the commonality between
- 8:59
all of these beautiful minds. And so this morning, started at ten AM at my favorite coffee shop in the [REDACTED:location], uh, with Opus 4.8 because Fable was not out at this point.
- 9:11
I created this really fun tool. I call it Shape of Minds, and it's a way for us to find commonality across different dimensions of the most beautiful brains in the history of humanity.
- 9:25
Um, and I can show you a little bit. It is, uh, highly responsive. You can just hover over
- 9:33
all of these beautiful minds, and it tells you what makes this person unique or in the dimension of thinking, for example, how was Bach, um, unique? And it's all, like, you can see the m- the spikiest minds, uh, based on these different dimensions.
- 9:50
So I wanted to explore their thinking, their obsessions, their quirks, their origins. You know, it's interesting 'cause here, for example, there are a lot of [REDACTED:origin] in these, uh, in these great minds.
- 10:04
Um, their routine, prolific output, workaholic, unsurprisingly, I guess. For love, interests, worldview, religion, habits, knowledge.
- 10:16
And because I didn't wanna bore you too much with all of this overload of information, I asked Fable to come up with the highlights, the most interesting things that it picked up on, uh, when it came to the commonality or spikiness of some of these minds across the different dimensions.
- 10:32
And there are really interesting things, and I'm gonna discover them with you because I haven't had time to read them before, uh, presenting here 'cause this was generated about twenty minutes ago, right before.
- 10:44
Um, so let's see. I guess people, these people like to har- to argue. Um, a lot of them were childless.
- 10:54
Uh, not sure what this means. Um, oh, yeah, naps. Naps were very popular among the best minds. Um, they barely ate. This is an interesting one.
- 11:08
Um, and yeah. Anyway, all of this is public, so if you wanna, uh, look at the highlights, it's, um,
- 11:17
it's quite soul nourishing and mind nourishing to look at this. Um, and I also created a design. So if you had not, um, picked up on this, the entire design language of this tool that I built this morning was in celebration of the Fable 5 campaign that came back.
- 11:33
Um, and so I asked, uh, Fable to extract all the butterflies from this and then turn them into individual components that I can use for the design language of this project.
- 11:47
Um, what else can I share? Um, oh, this is a MD of-- It's a glossary of all the information that is displayed on this project, super easily readable by the agent, and it's a lot more efficient than having the agent look at this page, obviously.
- 12:03
And funny enough, I made this page this morning when I started working on this project, and I knew that I had a-
- 12:09
A deadline to hit, and so I made a timer for myself, uh, when I started at ten, ten AM so that I knew exactly how much time I had to work on this project to accomplish the tasks that I needed to accomplish.
- 12:23
Um, and yeah, it's all like software on demand. Whatever, like whatever stream of consciousness you have, you should just ask an agent to do it for you.
- 12:32
Um, like all of this i- is AI. I generated this picture, which I kind of thought was cool. You can probably recognize some of these cool faces like Da Vinci, Salvador, uh, Dali, Marie Curie, Nikola Tesla, Einstein.
- 12:48
Um, and yeah, there are only-- I have five minutes left. What can I share? Um, I can share, [chuckles]
- 12:58
I can share-- So I love to make people smile. I love to make people smile when I can. And, um, a great way for me to make people smile is to use a lot of emojis, um, especially in the company Slack.
- 13:10
And, um, it was always a pain to find the emoji that I wanted to use to make people smile in the Slack. It was like very inefficient for me to find that.
- 13:18
So I created this tool for myself where I can just like search for a random, fun, stupid Slack emoji as a reaction, um, for my Slack. And same thing for these here.
- 13:30
They're kinda fun, specifically like this one. Um, and this is like the regular iOS emoji picker, which I kinda always struggle with. I never find the thing that I'm looking for, and so I just build it for myself, and you can like by a click just copy.
- 13:45
Um, another fun thing that I built is I always, you know, we're all students of life, and we're all trying to learn as much as possible. And with these, these new models, we can
- 13:57
learn at a pace that was never possible before. And so whenever you have a stream of consciousness about something that you'd like to work on, um, you should just ask it.
- 14:06
You should, you should just g- ask a, a model to prepare like a little ro- like report for you on this topic. And that's what I started doing a few, a few days ago.
- 14:14
All these things are things that I would have, that I love to, would love to deepen my understanding of, uh, but still make it very efficient. And so I just explored a few concepts like the ten principles of good design and, um,
- 14:30
like Renaissance patronage is also something that I'm fascinated by, uh, right now. And so it just cooked up like a short, very short report, uh, for me that I can digest as I'm on my way to work or I'm coming back from an event or something like that.
- 14:45
Um, that's why I'm so excited about cloud workspaces as well, is because it's gonna be very possible very soon with Conductor Cloud to
- 14:54
be able to spin off a bunch of agents on the go, um, as I'm like traveling from one place to another.
- 15:03
Um, but yeah, I feel like I can,
- 15:07
uh, I can stop there. I don't wanna overload you with a lot of things, but there's another interesting thing that I created. It's like this, the backlog of all the changes on this website as I was making, uh, the final tweaks to the shape of Mayan a few minutes ago, twenty-five minutes ago, and you can see like
- 15:23
the evolution of that repo over time. Um,
- 15:29
but yeah, I guess I can, I can end early, and if you have questions, I am happy to take it, uh, to take questions. All of this is an experiment, and I would love to learn from you guys on how you're experimenting with models.
- 15:43
Um, yeah. [clapping] [outro jingle]