# Englobing: the weekend my design fiction caught up with me

> A design fiction about a swarm remaking a charity's website meets Spawn, a company run by a fleet of agents, and a blind writer spends five days building games, walking Liverpool and asking the swarm's question for real.

Charli-Jo Tyrer · 3 November 2026

On Friday, I wrote a story about a blind woman sitting at her kitchen table, directing a swarm of AI agents to remake a charity’s website.

On Saturday morning, I heard a podcast about a company that was already working with something remarkably like that swarm.

By Sunday, I had commissioned a robotic guide, walked around a neighbourhood that hadn’t existed that morning, and returned to the kind of space trading game I last played when I could see.

By Monday, I was walking around Liverpool.

By Tuesday, I was stuck outside Bravissimo because my virtual mobility instructor thought I was still going to Liverpool Central.

There are probably more dignified ways to encounter the future.

Mine involved a bra shop.

But I should start with the feeling.

While I was creating the fiction, I reached for a word: *englobing*.

I had a sense of agents surrounding an existing website. Reading it, tracing it, finding its purposes, following its journeys, understanding the organisation it represented. Getting around the whole thing.

Then remaking it accessible.

I have spent enough of my life around accessibility reports to know what happens next in the conventional version of this story. Someone examines a website. They identify failures. They write them down. The report goes to somebody else. There are priorities, budgets, responsibilities and discussions about whether fixing the navigation will break the template.

Sometimes good work follows.

Sometimes the report becomes another inaccessible PDF.

What I was imagining went further. The capacity to observe and understand a service would sit alongside the capacity to reconstruct it. The people finding the problems would be able to direct changes, test the result and carry on.

The site would become something you could work on as a whole.

That was the feeling behind *englobing*. I didn’t yet know what I would be doing with it a few days later.

The fiction was called [*The Remaking of the Merrow Trust*](https://blind-ant.com/pieces/the-remaking-of-the-merrow-trust/).

Its protagonist, Ruth Ashworth, is a retired systems analyst who has been blind for forty years. She makes tea, sits at her kitchen table and wakes a swarm.

The Merrow Trust is an invented small charity providing training and advice for blind and partially sighted adults. Ruth gives the agents a carefully bounded brief. They may inspect the public website, reconstruct what they can establish, model the service and propose something better. They may not contact staff or start behaving as though the proposal has been agreed.

The observers come back with hundreds of pages, dozens of forms, duplicated PDFs and an email address attached to a mailbox that appears to belong to 2019.

Anyone who has worked with organisations will recognise the mailbox.

The analysts begin arguing about whether three differently named pages describe three services or one service with three histories.

Then an agent proposes retaining the existing navigation.

Ruth stops it.

The menu records how somebody once decided to file things. She wants to know what people come to the Trust to do.

Find work. Get help when their technology stops working. Understand what to do when they are losing their sight. Refer somebody. Give something back.

The work starts taking its shape from those intentions.

There is a line in the swarm’s configuration:

“modeller: Ruth Ashworth (human)”

She put it there herself.

That matters to the story. Observers, analysts, builders and testers can be multiplied. They can investigate many things at once. But somebody still has to hold a coherent account of what the service is for, what should happen and how anyone will know that it has happened.

Ruth walks through the journeys aloud.

Intention. Understanding. Authorisation. Action. Evidence. Correction.

Then, at 11:48, a tester announces that an application has been submitted successfully.

“Has it?” she asks.

The agent explains that the submit control returned without an error.

“That’s not what I asked.”

There is no reference number. No visible application state. No evidence that the charity has received anything.

The agent revises its report to *unverified*.

That was Friday’s fiction.

The following morning, I heard [*The AI Daily Brief*](https://aidailybrief.beehiiv.com/) discussing [Spawn](https://www.spawn.co/), a game-making platform founded by Jacob Sansbury.

On Thursday, 29 October, Sansbury had published an account called [*How I Built the Most AI-Pilled Company in the World*](https://twitter.com/jsnnsa/status/2115623382721126826). [Simon Willison](https://simonwillison.net/) wrote about it that day. The Friday podcast followed the same story: a small company using a persistent fleet of agents to build and maintain its product.

Willison had also tried Spawn on his phone. He prompted a castle into existence, then a dragon, then rode the dragon around the castle. His account included the company’s claim that agents read production problems and ship frequent repairs, alongside his reservations about applying that model to higher-stakes software. [Willison’s account](https://simonwillison.net/2026/Oct/29/how-i-built-the-most-ai-pilled-company-in-the-world/) and [the podcast notes](https://aidailybrief.beehiiv.com/p/how-the-most-ai-pilled-company-in-the-world-works/) are dated Thursday and Friday respectively.

I encountered that account on Saturday.

I’m keeping the order straight because it is what gets me.

The work was already happening elsewhere. I hadn’t somehow invented Spawn at my kitchen table. But I had reached for this particular picture of a swarm before hearing about this particular example of one.

I had written the relationship.

Then I found something with which I could begin trying it.

At first, the question was quite ordinary.

Could I use it?

I explained to Savi, Spawn’s game-making agent, that I was blind and used [NVDA](https://www.nvaccess.org/), the Windows screen reader. I wanted to explore a game world without depending on seeing its visual presentation.

The result was *The Keeper’s Cottage*, a small exploratory game with rooms and descriptions of what I would see as I entered them.

That was interesting.

On Sunday morning, I asked for something more ambitious.

I have been waiting for a robotic mobility device called [Glide](https://glidance.io/) since 2024. The user supplies the forward movement, while the device assists with steering and avoiding obstacles.

Could Savi create a game around that relationship?

I wanted to choose a destination, provide the movement and experience a guide taking me through a world.

Savi built Maple Street, in [Glide Walk](https://www.spawn.co/@charlijo/glide-walk).

There were houses, pavements, roads, pedestrian crossings, a park, a bakery, a café, a bus stop, a post box, traffic and a barking dog.

There were thirteen locations and three errands.

Post a letter. Buy a loaf. Sit beside a fountain.

It gave me instructions.

Hold W to push Glide.

I held W.

Nothing happened.

This is an important part of the story.

A screen reader can read an instruction perfectly clearly. The instruction can still be wrong.

Eventually, I tried the spacebar.

I could hear myself moving.

Footsteps. Traffic. A changing soundscape.

I chose the fountain in Willow Park, and Glide took me there.

I could hear the water.

I have spent something like 10,000 hours exploring artificial intelligence. I thought I had acquired a reasonably high threshold for being surprised by it.

Apparently not.

I was moving through a virtual environment I couldn’t see, using a mobility system I had commissioned by talking to an AI. The world, the interaction and the means of accessing it were being developed together.

I could try something, discover that it didn’t work and ask for it to change.

Then I asked for smart glasses.

If I pressed a key, could the game describe what was in front of me, as though I were wearing camera-equipped AI glasses?

Savi added the feature.

There were technical questions immediately. Was it interpreting a rendered image? Reading structured information about the world? Combining the two? Producing a plausible description from predefined data?

I hadn’t established the answer.

I could, however, request a description while travelling through the environment and begin investigating whether it made sense.

That distinction became part of the work. I could enjoy the experience while remaining precise about what it demonstrated.

A simulated robotic guide is a game mechanic. It tells us nothing reliable about the engineering or safety of the real Glide device.

A convincing description needs checking against the world it claims to describe.

And a game that lets me reach the bakery still has to let me buy the loaf.

Then I remembered [*Elite*](https://en.wikipedia.org/wiki/Elite_(video_game)).

I played it in 1984, the year before I lost my sight.

Trading. Navigation. Exploration. A galaxy in which to make decisions and see what happened.

I wondered whether I could create my own interstellar trading game.

A blind captain. An automated ship. A crew that could handle navigation, docking and the physical work while I decided where to go, what to buy and how much risk to take.

Savi built [*Blind Trader*](https://www.spawn.co/@charlijo/blind-trader).

My ship was called *Second Sight*.

I didn’t choose that name.

Its AI was called Lyra. There were two robots: Clunk, the cargo loader, and Pip, the flight and repair drone.

Lyra’s first greeting was:

“Welcome back to the helm, Captain. It’s been a while.”

Forty years, as it happened.

That line got me.

I bought twenty tonnes of liquor at Lave. Clunk loaded the hold. I chose Meridian, 3.8 light years away. Lyra plotted the course. Pip handled departure and docking.

At the other end, I sold the cargo.

I had started with 1,000 credits. I now had 1,339.

I kept playing.

I upgraded the cargo bay. Traded food and textiles. Bought computers. Refuelled. Made more money.

Eventually I had 2,008.4 credits and access to a wider chart.

An entire gameplay loop had survived repeated use.

There were inconsistencies. Lyra announced one number of systems while the interface showed another. Combat had been designed around turns and decisions, but I hadn’t systematically tested every part of it. There was still plenty to investigate about focus, announcements, persistence and the relationship between the rendered world and its accessible interface.

But I was playing.

I had wanted to return to a kind of game I remembered. I was now doing it in a form built around how I could participate.

Lyra recommends. I decide.

That relationship ran through both games.

In Glide Walk, the robot supplies steering. I supply movement and choose where we go.

In Blind Trader, the ship and crew supply navigation, docking and cargo handling. I make the commercial and tactical decisions.

The machines supply capabilities. I keep the decisions.

By Sunday evening, *Back to Elite* had become the title of an account of something I had actually done.

On Monday, I brought Glide to Liverpool, in [Glide Walk: Liverpool](https://www.spawn.co/@charlijo/glide-walk-liverpool).

I used to know Liverpool city centre well enough to walk around it. Liverpool ONE changed the geography I held in my head. Familiar places remained, but the relationships between them had shifted.

I wanted to start outside James Street station and go somewhere.

Liverpool Central, to begin with.

Then the Ship & Mitre.

I wanted streets, crossings, sounds, landmarks and choices. An instructor who could help me learn a route, and a solo mode in which I could find out whether I remembered it.

I also wanted heels.

If I was going to walk around my own virtual Liverpool, I could choose the sound of my footsteps.

Then I complained that the mobility instructor wasn’t talking to me enough.

Walking around was getting lonely.

So he became chatty.

These requests might look like incidental decoration in an account of accessible software development.

They were part of what made it somewhere I wanted to be.

The experience included hearing a saxophone on Church Street, recognising a place, listening to the instructor and choosing how I moved through the city. It included remembering Liverpool and enjoying a new version of it.

I wrote about it on [AppleVis](https://www.applevis.com/) and invited people to have a wander.

People did.

One person tried Spawn and decided they preferred playing and reviewing to creating. Another recognised an opportunity they wanted to take. Others were interested but needed time and energy to explore.

Those are all reasonable responses.

Access to a new way of making things doesn’t create an obligation to become a maker.

By Monday afternoon, there was also a San Francisco walk. Montgomery Street station, SoMa, imagined visits to Anthropic and OpenAI.

Geography was becoming rather optional.

I began considering Pompeii in AD 79 and London during the Blitz.

There is a particular danger in discovering that you can ask for somewhere and begin having it built.

You start asking for somewhere else.

That evening, I gave [Codex](https://openai.com/codex/) access to an [OpenStreetMap](https://www.openstreetmap.org/) service and asked it to make Liverpool more like Liverpool.

The arrangement became slightly dizzying.

I talked to Codex. Codex researched the streets and access information, opened Spawn and instructed Savi. Savi rebuilt parts of the game and ran checks in its own test player.

I listened, tried things and decided what should happen next.

The route to Central acquired real bends and a reported length of about 868 metres. The Ship & Mitre moved into its proper relationship with Dale Street, with approaches through Castle Street or North John Street.

The work also included checking what would happen to people’s saved games when the map changed.

Meanwhile, Claude was updating my website, and my iPhone told me when it had gone live.

Earlier, I had switched the air fryer on from an app and warmed something in my 28-year-old talking microwave.

The future was arriving alongside an appliance that had been getting on with its job for nearly three decades.

I wrote on AppleVis:

“I’m sitting here thinking: there has never been a better time to be blind. It’s still shit, mind.”

Some people preferred the mid-1990s. They pointed out that several of the problems AI is now helping solve hadn’t been invented then.

Fair enough.

My astonishment doesn’t require their experience to be wrong.

I was describing a freedom I had just exercised. I was also blind, with everything that still entails.

Both parts belonged in the sentence.

On Tuesday, Liverpool gained two more destinations: Apple and Bravissimo.

Apple meant modelling an upper level and a lift. Call it. Wait for the doors. Step in. Select a floor. Hear the announcement. Step out.

Bravissimo meant walking through School Lane and Peter’s Lane into St Peter’s Arcade, with Hotel Chocolat opposite.

The source information mattered. So did the limits.

A mapped shop point isn’t a verified entrance. A building outline doesn’t establish where the door is. The game’s lift position, walkway and shop interiors included approximations, and the text said so.

Inside Apple, I could speak to a fictional Specialist about VoiceOver.

Inside Bravissimo, I could collect an imaginary order and sit in the lounge.

No real purchases were involved.

The builder had been given a laptop upgrade, a bra collection and a chocolate shop to think about. I felt that was sufficient information.

Savi reported that the shops worked. It had tested journeys, the lift, shop actions and saved states.

It was also explicit about what it hadn’t done. It had checked text and logs. It hadn’t heard the announcements or tested with my screen reader.

We published so I could test the live game.

I reached Bravissimo.

The game described the entrance. Flat, double doors. It offered the way inside.

I chose it.

“Turning left, along the Bravissimo doors to Inside Bravissimo, 14 metres, heading east.”

Then the instructor interrupted.

“Not that way for Liverpool Central station entrance. Down arrow and Glide turns you round.”

I was outside a shop whose entrance the game had just offered me, being warned away because the instructor thought I was going somewhere else.

Has it?

The tester’s success report had covered a path through the implementation. My experience had uncovered a different combination of state.

The outside-shop destination, the instructor’s goal and the errand were not connected properly. After arriving, the instructor fell back to the first unfinished errand.

Liverpool Central.

I switched to solo mode.

I got inside, reached the collection desk and collected the order.

Bravissimo was recorded as done.

The game also charged me two wrong turns.

That needed attention too.

The problem reached into the model of the experience. The system had to understand that arriving outside Bravissimo and entering Bravissimo belonged to the same purpose. Walking between its door, desk and lounge should remain coherent. A previous goal should not take over at the threshold.

The repair connected those places to their errands, changed how the instructor chose its goal and adjusted the wrong-turn rules.

Savi reported tests for Apple and Bravissimo in instructor mode, with the shop interactions completed and no false wrong turns. It checked that actual wrong turns could still count, that a saved doorway state could recover and that the pub guidance still worked.

My completed Bravissimo errand was preserved.

The correction went live.

Ruth’s argument had become uncomfortably practical.

A successful control activation was insufficient. A destination announcement was insufficient. Even getting inside the shop was only part of the task.

Could I do what the experience promised, understand what happened and continue?

That was the question I had written into the fiction.

It was now the question I was asking from inside my own game.

I don’t want to turn a few days of experimentation into a claim that accessible software has been solved.

It hasn’t.

We encountered incorrect instructions, contradictory guidance, incomplete tests and a scoring error. There are still things I want to understand about the games, their sound, their visual worlds and their reliability.

Nor have I demonstrated Ruth’s entire swarm surrounding a charity’s service and reconstructing it into something validated by the people who run and use it.

That remains work to do.

But the feeling I called *englobing* has acquired substance.

I can see a relationship between observing a system, understanding its purposes, directing its reconstruction, experiencing the result and correcting it. Those activities can now happen much closer together.

And I can participate in all of them.

For much of my life, accessible software has arrived as something somebody else made, with decisions already taken about what I would be allowed to do.

During this weekend, I could originate the world, choose its mechanics, ask for the information I needed, encounter failures and direct repairs.

I could also ask for heels and a less lonely walk.

The professional questions are substantial. Who holds the model? What evidence establishes success? Which decisions belong to the person? What can the agents safely do? How do we correct them when their confident account disagrees with experience?

I had been working on those questions before I discovered Spawn.

Then there was the fountain.

And Lyra saying it had been a while.

And Liverpool, which I hadn’t seen for forty years, becoming somewhere I could begin wandering around again through sound, words, memory and imagination.

Those experiences are why the questions matter to me.

On Friday, Ruth sat at her kitchen table and asked a swarm what it had actually accomplished.

By Tuesday, I was asking much the same thing.

I had also made money trading between star systems, visited San Francisco and collected an imaginary bag from Bravissimo.

I have mostly been pissing about.

I have been doing some fairly serious thinking.

And I am still absolutely fucking astonished.

## Read more

- [*The Remaking of the Merrow Trust*](https://blind-ant.com/pieces/the-remaking-of-the-merrow-trust/), the fiction, at Blind Ant.
- [Blind Trader](https://blind-ant.com/pieces/blind-trader/), with its trailer, at Blind Ant.
- [Glide Walks: Liverpool](https://carrados.com/blind-ai/glide-walks-liverpool/), my first dispatch from the walk.