I went into my kid’s school a couple months back and spoke to the year group about manufacturing.

Honestly it was the most rewarding speaking gig I’ve done all year.

It was about the process of making my AI clock and I have a ton of pics from my factory visit to Shenzhen (mostly pics that I have only shared with Kickstarter backers).

I talked about where ideas come from and the value of playing around, and how it’s neat to learn new techniques that you can combine together.

I talked about prototyping and design – and was sure to use the words “prototyping” and “design”. I showed exploratory sketches and what CAD looks like.

I handed round various iterations of e-paper screens, and electronics from breadboard to PCB, and various iterations of plastic parts.

It’s interesting to see how a plastic enclosure comes apart, and to connect that to what an injection moulding machine is doing.

(A lot of the kids are familiar with 3D printers, so I showed a timelapse of a 3D print – it would take a year to print all my clocks! And then a real-time video of injection moulding, and how that would only take a day.)

And then photos of factory floors, and here’s the team, and assembly lines and what a page from an assembly procedure looks like, and packaging too.

7 year olds have great questions.

Like: how does it not break in the post?

Well here’s a vibration machine in action and that’s how we test it.

And, look, in this cardboard packaging, here’s a cradle, and this was made by a packaging designer – you could be a packaging designer too if you want.

Like: how does the button work?

Well you’re right I didn’t pass round the separate button piece, good spot, it’s small and I didn’t want to lose it. So now let’s talk about assembly and about industrial designers…

I don’t like those videos of factories that are supposed to inspire awe.

You know the ones I mean: you see a thousand products a second whizzing by on 20 parallel belts. You come away saying wow. When they showed manufacturing on kid’s TV when I was growing up, that was what they showed.

“Awe” is the opposite of what I want to convey.

Except for a very specific types of person, when you show something with the expectation that “awe” is the appropriate response, you are implicitly saying to your audience: you should step back here and appreciate this from a distance. Like looking at a great work of art. Gasp but do not place yourself in the picture.

Whereas!

I want to re-home manufacturing. I want these kids to become designers, engineers, inventors, factory owners, and all the rest. Makers of any kind; participants in the ongoing making of our world.

So my message is: sure this is complicated but it’s fine, we can do complicated.

Factories are just rooms.

The stuff around us isn’t divine - these chairs we’re sitting on, the TV at the front of the classroom, the pots for the plants - all this stuff was invented and figured out and made by people.

p.s. you can be one of those people.

So when I heard the class was learning about inventing, I offered to go in and show that scrappy dead ends are cool actually (it was amazing to speak with a class that already knows the word “prototyping”) and this is electronics and this is going from sketching to plastic and this is what it means to make a product and to sell it.

I deeply feel this mission to normalise getting our hands dirty with the world – when they’re 7 years old, while their brains are still establishing what’s normal.

(This is connected with what I was saying about training for collective efficacy.)

And I’m just someone’s dad, you know? So if this guy can do it…

If you have the opportunity to go into your local school and talk about making things too, please do. You will be rewarded with wonderful curiosity, engagement and questions from the kids.

Hopefully one of them one day will look around them, think “someone should do something about that”, remember back, and say - oh that someone can be ME.

More posts tagged: that-ai-clock-and-so-on (15).

Auto-detected kinda similar posts:

• A fourth one-off never-to-be-etc Acts Not Facts weeknote (27 Nov 2023)
• Hallucination should not be a dirty word (26 Jul 2024)
• My top posts in 2024 (30 Dec 2024)
• Two upcoming talks on AI (5 Jun 2025)
• New interview (16 Dec 2015)

1.

It’s hot in London so I’ve been seeing a lot of handheld portable fans, usually with a strap so you can hang it round your neck.

My faves are the ones with thermoelectric coolers in the middle of the fan: a small plate that is so cold that it is covered with icy condensation. It’s the Peltier effect (the plate is really hot on the back) and the first time I’ve seen a thermoelectric cooler in the wild.

Here’s a thermoelectric handheld fan on Alibaba. Three quid each if you’re buying over a million.

A fan looks like a fan because of the mechanism used for the movement of air.

I’ve been meditating this week on what motivates form in product design.

Hey free concept: AirPods with built-in Peltier thermoelectric coolers so the buds are ice-cold in your ear holes.

2.

A fan moves air; electronic products move data. That can motivate the form just the same.

Durrell Bishop’s Marble Answer Machine (watch the video):

marbles dropping out of an answering machine could form an intuitive physical interface. This work later became the seed for a new movement called Tangible User Interfaces.

Each marble is a message that you can place in the player dish, put aside to keep for later, and so on. So sophisticated and so immediately understandable. (“Legible” as Durrell say.)

But what about when a product can do anything? Like a phone?

We end up with anonymous slabs of black glass.

3.

Before movie theatres there was the Kinetoscope (Wikipedia) "an early motion picture exhibition device, designed for films to be viewed by one person at a time through a peephole viewer window."

The Kinetoscope came out of Thomas Edison’s lab and established the idea of reels of film, and also film as “content” to be manufactured and distributed.

But it was a single-viewer device: you leant down and put your eyes to the viewer.

The act of peeping – the form is motivated by the human interaction.

Or for the necessity of the affordance: a possible interaction that must been seen as a possible interaction. (Affordances as previously discussed.)

BTW:

I recently discovered via the sf journal [Foundation] (issue 152) that:

Thomas Edison, using his Kinetoscope, is credited with producing (though not directing) … the first filmed sneeze (1894), the first filmed kiss (1896).

4.

The egg timer that looks like an egg is the best product design of all time.

Here’s one on Amazon.

The traditional kind of hourglass timer that uses sand, on the other hand, is rubbish:

• It is traditional
• Its form is motivated by the mechanism
• And it is legible because you can see the passing of time.

But what to do you use it for?

Measuring time, sure. A lot of stuff. It can do anything (related to waiting for a period of time).

But does an initial use come to mind?

You have to think about it – aha eggs! Or you have to learn it. Ultimate the sand timer is abstract.

Like an empty ChatGPT window?

AI can do anything too.

But I opened my ChatGPT just now, and look how hard they work to give you ideas of what to do. Mine suggests: Write an email, create a painting, give me ideas…

If the hourglass timer were designed like that, they would print on the side:

• For eggs
• For rice
• For pasta
• For reminding me how long I have till my program is on.

So cumbersome.

Whereas!

The egg timer that is shaped like an egg.

Form follows function – but also where and when and how.

And then once you have timed your eggs, you have in your mind this new hammer of “timing” and you see immediately everything else you can time.

So another motivation for form is to imply the first context of use, even if - and especially if - the product can be re-purposed or adapted for other contexts by the end user, once they have internalised the function of it.

It is genius. I aspire to design a product this perfect.

The egg timer shaped like an egg was invented and patented by Lucio Oliveri in 1982. U.S. Design Patent No. D276,705 (expired).

More posts tagged: filtered-for (124).

I didn’t know what a near-infrared LED mask was a week ago, and now I am obsessed with this ad from Omnilux.

Because there’s a lot going on.

1. The aesthetic of the mask itself is wild: the lines look like sinews as if the woman has been flensed
2. But the schematic flesh is commercial! The logo on the forehead!
3. Yet she has glossy hair and a spa outfit
4. And is drinking a glass of white wine??
5. Let’s not get into the red glow through the holes in the face
6. The ad copy is tight af, no messing around: retail, popularity, timely, professional, boom boom boom.

"Your best skin awaits"

I have questions about the very concept of an LED mask (can’t you go outside? but now we’ve internalised a fear of the sky because of UV?) but that’s not my point.

These components don’t work together. Health and beauty vs the evil red glow. You can’t sip the wine through that mask.

So there’s no singular integrated vibe here. It’s the opposite of vibe. It’s a set of hieroglyphs. Six symbols collaged together into the same image.

Similarly the vacuum guy James Dyson is now growing strawberries and the copy is wild.

"AI-powered British strawberries"

I mean let’s unpack that just for a second…

• “AI” is high tech future-y but also people are trained into not asking questions about it, just that it must be great because trillion dollar companies are here
• “British” has a surface of local product but in a Dyson context is elbows-out national pride, pro-Brexit style (as he is, even though he moved production to Singapore)
• “Strawberries” as in strawberries and cream and Wimbledon.

Let’s not get into the photo in which we are reassured about the quality of the strawberries not because they are being nurtured by a friendly farmhand – but because they are being CCTV monitored? Like: AI-powered panoptic strawberry surveillance will scare the strawbs into being plump and red, Jeremy Bentham’s paranoia-based fruit production?

The very next section is titled "British strawberries: 100% of your daily vitamin C" which is a whiplash into health.

At this point I don’t feel like I’m reading. I feel like these ads are laser-targeted streams of signifiers treating my psyche as a combination lock to be picked.

The fact that the grab-bag of symbols appears to have meaning on a human level (a photo of a woman; strawberries growing) is almost an accident. But there’s no content there beyond that.

Umberto Eco, semiotician, would have been able to write 2,000 words unfolding that Omnilux advert into its constituent symbols and deducing the shape of society from its very existence.

Peak semiotics was probably, what, the 1970s? We need that expertise to dismantle communication once again.

Anyway this is what it must feel like to computers when they get hacked. Like by a text message with a weird collection of characters that buffer overflows and takes over the app executable.

When AI gets really good - probably not much better than today - it will be able to automate the process of discovering the four or five symbols that unlock the “I must buy that” response, and then it’ll wrap it in a jpeg and put it on a billboard.

And next thing you know you walk by a photo of a woman drinking wine and it’s a jumble of symbols and you’ve been jailbroken and the compulsion to try an LED mask is so potent because well, I wonder what near-infrared feels like on the skin and what was that Omnilux you say? and without really thinking your phone is in your hand and

Auto-detected kinda similar posts:

• An Ariston TV spot from the early 1990s (23 Oct 2020)
• The Prompt Whisperer (4 Aug 2022)
• Viscerally and deliberately unsettling product design (14 Jan 2021)

Wet lab.

“Wet lab” was a new term to me, I had to look it up – it’s a laboratory with chemicals where there’s spill risk etc. I have a physics background myself and the closest I came to liquids was liquid nitrogen, and when you spill that it just boils furiously and goes away.

Don’t do this: if you put your open hand into a dewar of liquid nitrogen, you can hold it under the liquid at -196C/-320F because the boiling on the surface of your hand makes an insulating quilt. It’s a fun game. Closing your hand will create traps and you’ll get horrible burns.

Wet market.

The Wuhan wet markets entered the vernacular during Covid right? There are new things you need to refer to in a new context.

Wet-bulb temperature.

"The lowest temperature that can be reached by the evaporation of water" (Wikipedia).

It’s a function of air temperature and humidity and humans can’t survive about a wet-bulb temperature of 35C, equivalent to 40C with a relative humidity of 75%. i.e. sweating is no longer effective so you heat up and your proteins denature.

I first encountered this term in The Ministry for the Future (2020) by Kim Stanley Robinson which traces the worsening climate crisis, and wow, there’s a new term for a new context that it feels like we will be seeing in the news more in the coming years.

Wet pub.

A pub that does not serve food.

Wet signature.

It’s like “landline:” we never needed a word for a phone that was tied to the wall before there were phones that weren’t tied to the wall.

We never needed a term to say that this signature was made by an actual present human before we had photocopies and then computers.

Although somehow a robotic, automated autopen is granted “wet signature” status?

Autopens as previously discussed, which also mentions Margaret Atwood’s Longpen which is not automated and requires her to be syncronously present – but perhaps a 1,000 miles away from where the signature happens.

Wet hire, wet lease.

Equipment hire that includes human operators is wet hire.

As an aircraft operator, leasing a plane that includes staff is a wet lease.

Wet implies humans in both cases.

Wetware.

Hardware, software, wetware.

Back in the 80s, wetware - used somewhat ironically - meant your brain: “thinking” is running a task on the 3lb of wet meat in your skull.

I get the feeling that wetware today means lab-grown neurons? Wetware computers that run on cerebral organoids.

We didn’t really have words that weren’t made by humans until recently. Except for divine words and random words, both exalted.

I suppose we had widely reproduced words, and we call that “copy.”

But from now on, I guess, most words will be words not by humans, and that’s the new default. So we’ll need a name to specifically mean human words.

Wet words?

These are wet words!

One day we’ll have cyborg sleeves that move our arms for us and we’ll need a word for our own movements, and brain-computer interfaces that inject ideas and unearth memories and do reasoning for us, and we’ll need to call out plain old human brain-work: look ma I thunk it all myself.

Wet thoughts.

When we end up voluntarily ceding control of our own body parts to AI (for example so we can automatically walk around the grocery story while simultaneously catching up on boxsets on a VR headset) I wonder what that will feel like.

The underlying tech is currently in research phase.

What I mean is that humanoid robots are coming (as previously discussed).

And while robots are a mechanical challenge and a software challenge, some people are asking: what if you took that software part and pointed it at something else? Like, not a robot shaped like the human body but the human body itself.

For example: you can directly drive the muscles using Electrical Muscle Stimulation (EMS).

Here’s a new wearable called Human Operator which won MIT Hard Mode 2026.

Human Operator is a human augmentation tool that allows AI to briefly take control of your body to help you learn and do things you normally cannot do. To do this, it uses a Vision-Language Model for human motor control through Electrical Muscle Stimulation (EMS). Vision-based commands are generated via open-ended speech input through the Claude API to control finger and wrist stimulation for intuitive on-body interaction.

So you wear this augmentation over your arm. It looks like white fish scales.

Then you talk to it. And it zaps your muscles to play the piano: "AI stimulates fingers in sequence to play melody."

That’s the link to the GitHub page in case you want to build it yourself.

Cognitively, zapping your muscles to perform a task frees up your brain to do other stuff!

Here’s a paper that studies this:

Nith, R., Ho, Y., & Lopes, P. (2024). SplitBody: Reducing mental workload while multitasking via muscle stimulation. In Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ‘24) Article 81, 1-11. Association for Computing Machinery.

The authors point out that mostly EMS is used for tasks the subject is already focused on, e.g. "playing a musical instrument."

And also EMS is distracting, either because it "causes a tingling" or generally it will "decrease the user’s sense of agency".

But despite that, would it be possible to perform two sub-tasks, such as:

continuously stirring the pot to make caramel (a repetitive muscle movement) and writing an essay (a cognitively-demanding task).

The researchers called their EMS implementation SplitBody. It looks like a cigarette packet strapped to your upper arm.

Results:

We found that with SplitBody, participants reported less physical-demand (decrease of 31%) and less mental-demand (decrease of 26%) than when performing the task by themselves. Moreover, the performance increased by 35% (averaged over both tasks), including the task that was not automated by EMS, which increased by 18%.

i.e. if you’re on a work call and cooking at the same time, you should let yourself get electrocuted to do the cooking because, although it tingles, you’ll end up doing better at the work call and the cooking too.

Multi-tasking!

Electrical Muscle Stimulation is step one.

You could go direct to the brain?

Like, you can have robot cockroaches:

There’s a backpack that you can install on cockroaches that drives them around by zapping the neurons in their antennas with electrical pulses.

It was a Kickstarter in 2013: The RoboRoach: Control a living insect from your smartphone!

There’s also a technique called transcranial magnetic stimulation that you can use with humans: you zap your brain with a whopping great big magnet, and if you point it at your motor cortex then it can move your arms and legs.

Apparently it feels like free will.

Every so often I dream of a transcranial magnetic stimulation helmet that can walk me to the shops (2015):

That way your can check your email and Skype your mum, while the Walking-Down-The-Street Hat takes care of the tedious job of moving your legs, and collision-detecting your way around obstacles like buses and humans.

Whenever I meet someone who I reckon could genuinely build me a TMS helmet (I met such a person the other month at a lab in New York) I pitch the idea so it lives in their head.

And it turns out that, when you propose a special hat that automates your legs so you can walk to work while you’re also doomscrolling or whatever, they look at you like you’re loopy. Especially if you also act it out.

We have this image of ourselves as being singular, right? We do one thing st a time.

Yet I don’t consciously think about my legs when I stand up and walk, and I’m clearly multi-threaded in my thinking otherwise ideas wouldn’t develop in the background and pop up in the shower later, and I delegate my way-finding cognition to Google Maps and my memory to my notes.

It’s just an extension of that.

But I wonder whether it will feel alarmingly alienating, like voluntary disassociation as soon as you hit that button of the app that walks your legs to the office (it will inevitably use an app); or will it feel like being in the back of a cab where you can chat with your friends instead of thinking about driving?

And, as we get accustomed to running multiple embodied tasks in parallel, will we start feeling comfortable with renting our parts of ourselves on demand, just casual you know, and somehow it won’t undermine our sense of self?

It’s not so much that I think that a bodily-automation helmet is a good idea, it’s that it feels like an economic inevitability – once possible.

For example:

• Would you spend the price of a car on a home robot that can do the washing up and fold the laundry and tidy the kids room? Instead you could spend 10% of the money on an AI sleeve that takes over your arms to be the robot, and you can delegate your zombie limbs to do the chores while also attending your work zoom or winding down by doing your puzzles.
• Would you accept wearing AI gauntlets to drive your car if it meant lower insurance?
• If I’m sitting around just watching TV, why not wear an Apple Vision Pro so I can carry on watching my programmes on the inside, and meanwhile people on the other side of the world can dial into my body and go touristing around my neighbourhood, experiencing it from their headsets? I wouldn’t even have to know.
• Will a factory really bother investing in precision robotics to assemble their parts, or even time in training people to do the same, when they can strap workers with dextrous fingers into pre-trained software to do the job immediately – and the workers accept less pay because they are doing double-duty by working second tier call centre support jobs at the very same time.

Ok horrific.

But what extra could you do too?

Like, I wouldn’t trust myself to free-climb up El Capitan.

But maybe Electrical Muscle Stimulation could drive me up there safely, and I’d just be along for the ride as I (or whatever split body “I” am at this point) ropelessly ascend 3,000ft of sheer rock face, AI controlled hands and feet, high up in the fresh air of Yosemite, body tingling as a thousand tiny electric pulses zap me hour after hour through my fish-scaled cyborg bodysuit.

Auto-detected kinda similar posts:

• Cyborgs vs rooms, two visions for the future of computing (13 Oct 2025)
• These robot legs are made for walkin’ me to the shops (4 Nov 2022)
• Cyborg prosthetics for limbs that don’t exist (7 Apr 2020)
• Filtered for home robots, fast and slow (20 Sep 2024)
• Filtered for musical cyborgs (29 May 2020)

Ok a bunch of friends have their own personal AI agent using OpenClaw or Hermes or whatever.

Confession time: I don’t.

I haven’t found I need one. I’m a heavy user of agents for coding, sure, and a few isolated tasks. I haven’t found I need one digging through my notes or sending emails for me.

But really because of two fears:

• Data security. I’m cautious of something that can non-deterministically execute and get tricked into emailing confidential info or accidentally delete old photos without me noticing.
• Psychic integrity – I’ll say more about this in a bit.

So I’m a scaredy cat, whatever. Those concerns have kept me away.

Well, you know, concerns go out the window when you have your back against the wall? Last weekend, that’s what happened.

I’m shipping clocks.

Yes, finally, I know. It’s been a journey. I’m really grateful for the support and the patience of all my backers.

But, you know, 700 have left the warehouse in Hong Kong, and 640 or so have been delivered. The rest are in transit.

And last weekend…

Well, I’d shipped out 120 or so mainly to the UK and the US and everything had gone fine, very smooth.

I think I had two or three queries from US customs about import clearance and they just wanted a bit more information about the package. I filled in an online web form with a longer description than on the commercial invoice.

Shipping paused because the warehouse ran out of bubblewrap or something. That’s on me.

Then the next 500+ went out and I woke up on Saturday morning to 15 requests from FedEx for extra docs for customs clearance.

Ok that’s fine, I thought, percentage-wise this is approx the same.

So I started going through the requests. It wasn’t the simple text fields on the web this time. FedEx needed a worksheet. A worksheet is a detailed PDF that you have to complete, and they decide whether import is allowed based on that information.

The clock worksheet is not so bad. But it needs to have the tracking number at the top of the PDF, so it’s a different file for every package.

You go to the FedEx website which is arcane. Then you click around, upload the PDF, and you wait a few days and then the package clears customs and is on its way.

While I’m doing these 15 worksheets, which was the expected number, another 15 emails arrived.

I always do my best to embody the three virtues of a great programmer which are, according to Larry Wall

• Laziness
• Impatience
• Hubris.

So I figured that although it wasn’t taking too long to edit these forms, adding the tracking number to each, it would be appropriately lazy to spend essentially the same amount of time (a little more actually) to make a Claude Code skill where I type /toolbox:make-clock-worksheet <tracking> and the PDF is created and opened on my desktop so I can drag and drop it into the website.

SIDE NOTE: I have a private repo on GitHub which is my personal Claude Code plug-in called toolbox. It has all my custom skills and I can sync them between my machines and I improve them over time. Everyone should do this.

I woke up the next morning and another hundred queries had arrived.

I don’t know what kind of flag had gone off in FedEx internal systems but I think they believe I am smuggling watches or something. Because all the questions about my clock are like, ok so what kind of movement does your so-called “clock” have and how big is the movement (btw it’s not a clock unless it’s bigger than X by Y) and yeah so this “clock” how many jewels does it have.

Jewels? My clock has no jewels. Watches often have jewels, right? Watches are a great way to plausibly move value across borders. No kidding the Swiss have private banks and premium watches, a coincidence I’m sure.

(I have since been informed that jewels in watches are actually tiny and used mechanically as bearings so (a) TIL and (b) apologies for unnecessarily disparaging Switzerland.)

Anyway so I have to officially declare that these packages are not watches, they’re clocks, but I have a hundred of these things to fill in.

It is at this point that I realise what a freight forwarder is for.

But it’s just me, and I’ve got to get through these forms, and it’s a long holiday weekend, and I am seriously not going to be grinding through FedEx’s website for the next two days.

I gave in, I had to:

Claude Code, here you go, have my Gmail.

Claude Code, here you go, drive a web browser without me even looking.

Then I built an agent to process my email, navigating the FedEx website and uploading worksheets for me when that was the request (all the requests are subtly different), replying to emails from humans and attaching the worksheet when that was what had to happen.

I started conservatively, human-in-the-loop as you’re supposed to do, centaur-style, checking the drafts, but there isn’t the time. I admit it. I surrendered.

And then Monday, the holiday day itself, was pleasant in the end.

My family left for a few days. I sat in the garden in the sun washing down the bbq keeping half an eye on my agent as it worked its way through my inbox autonomously. What a life! Idyllic.

What I hadn’t quite internalised about personal agents is that they’re slow.

So time-elapsed is more but time-attended is lower.

It’s an odd kind of efficiency and doubly relaxing.

So it turns out I have a personal AI agent now.

Or at least the beginnings of the kind I am ok with.

Although it can send email on my behalf, which risks my data, I haven’t crossed the Rubicon to that psychic risk I mentioned before which would be allowing it to read and write into my notes.

My notes.

Let me tell you my theory about AI psychosis.

A lot of people keep a lot of notes.

I keep a lot of notes too, that’s how I write this blog, and in particular I like the serendipity of running across old ideas in my own notes – that’s common for other people too (2021).

We used to call it having an outboard brain and it’s true, I think for a certain kind of person, your notes become part of your extended cognition, and you “know something” whether that knowledge is within your skull or within your notes, same same, it’s just a matter of look-up latency.

My theory is that allowing something else to write into your notes does something bad to your psyche.

I had a glimpse of this: a few years ago I asked ChatGPT to write a blog post in my style. (This was before chat could browse the web; my blog is well represented in the training data.)

It was pretty good so I pasted it into my notes as a record (but never posted it of course). I got scared off using ChatGPT to help with my blog pretty early when I was talking through an editing decision and it came up with a turn of phrase that was so perfect and so unique that I couldn’t resist it. But it didn’t represent any thinking that I had done to arrive at it, this perfect metaphor, so it wouldn’t bear my weight when I leant on it. Those two experiences terrified me.

Anyway so recently I was browsing my drafts folder and I ran across the bottom half of this fake blog post without noticing the context at the top, and it was like when the elevator drops faster than you’re expecting because I read these words but they didn’t feel buttressed with even a glimmer of memory in my head, so I was gaslighting myself – had I really written that note? I mean there it is, it sounds like me, but I can’t think around those words.

The feeling of not being able to trust the permanence and integrity of the physical world around you is one thing.

Not being able to trust what’s going on in your own mind is another.

Am I the same person as I was yesterday?

So unnerving.

All of which to say is that, for me, my personal theory is that AI psychosis comes from undermining your intrinsic faith in the workings of your own self.

And that comes from allowing an LLM that speaks in your voice to potentially write into your notes which, for a certain kind of person, is part of cognition itself. The AI doesn’t need to actually change your notes, the potential is enough.

Which eventually makes you go loopy.

So I have this fear of risking my own psychic integrity, which has so far kept me away from allowing a personal agent to run on my own machine – I love automation but at a healthy arm’s length…

Yet needs must. And here we are. All it took was 100 emails requesting 100 custom PDFs.

I’ve opened the door and now it’s the slippery slope. The slope ends with outsourcing consciousness itself, and the prospect of humanity becoming a population of voluntary p-zombies is a high-probability eschaton that I’ve talked about before.

I’m making a note here to

• record my feelings at this moment: how naive will this look in a year or two?
• celebrate 640 or so clocks in people’s homes, more in transit, and 300 online right this second.

For both: thank you FedEx.

More posts tagged: eschatology (10), that-ai-clock-and-so-on (15).

Auto-detected kinda similar posts:

• Headless everything for personal AI (18 Apr 2026)
• My personal AI research agenda, mid 2024 (and a pitch for work) (7 Jun 2024)
• Who will build new search engines for new personal AI agents? (20 Mar 2024)
• The natural home for AI agents is your Reminders app (15 Jan 2026)
• Me talking about AI elsewhere (14 Aug 2025)

We’re open sourcing Resident, our library for running AI-authored code on microcontrollers – with no compile step and no firmware flashing.

It’s our twist on vibe coding firmware, built for instantly loading new device functionality coded by end users. It’s aimed at device developers, like us. We use Resident in all our work.

(We = Inanimate.)

Resident gives you a code sandbox on ESP32 devices and a driver API to provide hardware control and events. So an end user can push an app over Wi-Fi that instantly turns their clock into an interactive pill timer (for example) but the app can’t run probes on the home Wi-Fi network.

It comes bundled with a set of Claude skills. I told Claude about the capabilities of a compatible dev kit and asked it to push a simple app. Here it is:

At Inanimate, we believe that on-device sandboxes are an essential low-level primitive for AI agents in the real world.

The announcement is over on Lab Notes but I want to take a moment to connect this to some previous themes…

Look, I want to bring software into my room.

Like: if I’m working with Claude Code and I step away from my keyboard, it should be able to give me updates on what it’s working on by taking over my desk clock, and ping its permission requests to the gumstick device I have in my pocket for me to accept/decline while I’m making tea. Untether me from my desk! (btw I built this, it was awesome.)

Or - more prosaically - why can’t I yell at my stove as I’m leaving the house in the morning oh I forgot I’m roasting a chicken that needs to be ready for 7pm and have it look up timings and push me a notification when it’s pre-heating, and show a custom basting timer app on its 14-seg LED display.

"Why can’t I point at a lamp and say ‘on’ and the light come on?" – I was asking in 2020.

So half of the solution here is AI: LLMs are super good at translating intent into action. I use regular language and the computer will Do What I Mean (2025).

But the other half is a problem: how does this actually work? Where does the software run?

Every device needs to be able to run user code, that’s the answer.

I’ve been bouncing off this since 2023 when I put an LLM in charge of a smart home (using tech that we now call agents, but then was my implementation of the ReAct pattern from a paper out of Princeton and Google Research).

In a nutshell, let’s say you press a button on your desk clock and your AI agent has to decide in the moment what to do with that. It won’t work, it’s too slow. Cognitively, an interface has to respond inside 150ms or it is no longer instant – but the AI is a network hop away.

Ok so let’s remove the network hop. Let’s pretend we have edge AI: a GPT-4-equiv AI as a component in every physical thing; ubiquitous intelligence is coming and "a Feynmann-level light switch could guess your intentions pretty well," that was was my guess in 2023.

It’s not an outrageous extrapolation! Taalas is baking LLMs into silicon and delivers "17k tokens per second per user on Llama 3.1 8B" (try it here, e.g. 4,000 words on Hamlet as a space opera, it’s wild it’s so instant). So GPT-4-equiv is a matter of time.

Even with that speed, it turns that inference is not enough. To Do What I Mean, you still need to import user context for personalisation and grounding. But memory, Gmail, Wikipedia and the rest are still mostly in the cloud. That network hop again.

So AI can’t be inside the event loop, not if you want really great on-device interactions.

Instead – let the AI write device code. That’s the approach we’ve found.

Take that toy example of a roast chicken basting timer on the display of my stove. My AI agent should be able to dynamically write code for that interactive app, and have that app code executed on the stove itself.

I’m not saying that the AI needs to vibe-code firmware.

Firmware is the code that runs on device microcontrollers: you author it and compile it and you flash it and from that point on, it never changes. AI agents can vibe firmware (very happily). But I’m not sure I want to load code onto my stove that has boundless control over the heating element and the network stack and whatever other low-level capabilities are managed by the firmware itself. We’ve already had smart fridges sending spam email (BBC News, 2014), no more thanks.

Instead of firmware, the AI can run code in sandboxes.

Cloudflare made the case for sandboxes in March this year when they introduced dynamic workers:

Last September we introduced Code Mode, the idea that agents should perform tasks not by making tool calls, but instead by writing code …

You can’t just eval() AI-generated code directly in your app: a malicious user could trivially prompt the AI to inject vulnerabilities.

You need a sandbox: a place to execute code that is isolated from your application and from the rest of the world, except for the specific capabilities the code is meant to access.

Sandboxing is a hot topic in the AI industry …

We asked ourselves:

What if, when we allow an AI agent to spread its arms and stretch its legs in a room, it could inhabit devices by writing code that runs in an on-device sandbox, a sandbox that gives access to buttons and screens but not the network stack, and the app code could load and run instantly?

That’s Resident. It provides a code sandbox for ESP32 devices, and a toolchain for AI agents to write apps that target that sandbox.

At Inanimate, we use Resident for all our product prototyping.

And it will be at the heart of our future products.

Look, it may seem wildly disproportionate to write code to turn on a lamp.

But what are computers for? They do the hard work to make it easy for us. So this approach scales well from basic on/off control to… well, it turns out that, now we have this sandbox, we can compose all kinds of useful experiences that take over an entire room – and also weird and wonderful interactive ones that still work when you pull the network cable.

(An app arrives over the network but then the network is no longer required. Resident is built on our messaging library Courier which includes UDP multicast for local inter-device messaging even when internet connectivity drops.)

We’re opening Resident today as an alpha (v0.5.0) and open sourcing it under the highly permissive MIT license. Just include our copyright notice with any modifications.

Technically, we’re adding a Lua runtime to your ESP32 device. (Lua is a language designed to be embedded.)

We love Espressif’s ESP32 microcontroller family because it has a unique span in the ecosystem: it is used by individual makers, new hardware startups, and in production at real scale. It has built-in Wi-Fi and supports its native framework esp-idf and Arduino too, which is great for quick prototyping.

Resident gives you an API to add extensions to that Lua runtime: hardware drivers. Those managed capabilities are what makes it a sandbox. So the apps in the sandbox can respond to events that your button driver injects, and they can write to the display via a module added by your display driver.

The apps can be hot loaded at runtime. The way we have it wired up, you push app code down a websocket to the device and it run immediately in the sandbox.

Adding Resident is straightforward during development: bring up your new device as normal, then point your coding agent at docs/start-building.md. It’ll walk you through adding the sandbox and writing the drivers.

Resident comes bundled with:

• Connectivity: websockets, JSON messaging, and easy Wi-Fi config
• A default back-end server at resident.inanimate.tech so you can easily push new apps and events – use the example code to build out your own back-end when you’re ready
• A collection of Claude skills to create, validate and push new apps to your devices (and even write device documentation)
• Example projects.

Want to try Resident now?

Ahead of developing your new device, pick up an M5StickS3. These are made by M5, I talked about them when we announced Courier.

The M5Stick is a dev kit that has an ESP32 with a screen, battery, couple of buttons, buzzer and IMU. You can bring it up using Arduino pretty simply. Then add Resident and start writing apps. We have an example project.

Want to get going even faster?

You don’t even need hardware…

The Try it now section of the Resident homepage has a M5Stick simulator, running the Resident sandbox in-browser.

Drag and drop an app onto the simulator to see it run.

Or even: install the Claude skills, tap the button on the webpage to make the simulator live, and create apps from your local Claude Code session. The in-browser simulator will update live and run your app.

Hey, deep cut reference alert:

Back in 2021 I went on a dive into files: Golems, smart objects, and the file metaphor.

What would it mean to drag and drop a file onto a lightbulb? I asked. "Do I literally mean that the lightbulb needs a little slot like the golem’s mouth, into which you insert your instructions stamped on microfiche?"

This is my answer haha

Resident is what we’ve been using to prototype products and use cases at Inanimate.

We believe that one day all products will work this way.

For more info, the GitHub, and to try it now: Resident.

For updates: subscribe to Lab Notes.

More posts tagged: inanimate (6).

1.

I love this terrible fake address generator:

We have spent 3 years collecting data from every country to generate this enormous database.

…it promises.

These are not English addresses:

"2 Scott Common Jenniferstad S60 2PT"

"4 Matilda Via North Emilychester M20 1BT"

"Flat 48v Harrison Motorway Lake Gary EH4 5LQ"

Good story: Inside Number 0, John Finnemore.

2.

Kentucky Superstitions (1920) by Daniel Lindsey Thomas and Lucy Blayney Thomas:

Some household and domestic life superstitions, from part 4:

• "An accidental dropping of a knife brings a woman caller."

(My nan would always say it would be a man caller, to the point that if I drop a knife I reflexively say “man coming” to this day.)

• "If you spill salt on the table, you will get a scolding before Friday."
• "If you burn salt on the stove or in the grate, you will have to pick each grain out of fire in hell."

Salt matters eh.

Part 3:

Some of these I recognise…

• "If you raise an umbrella in the house, you will have bad luck."

Some are new to me:

• "If you whistle in bed, you will cry before you retire again."

Some are quite good, so have a read. (Some are horrifically racist.)

3.

A psychology paper from 2019 categorises 28 enjoyable emotions:

The families of discrete enjoyable emotions, many proposed for the first time, are as follows: (1) Self-praising emotions (authentic pride, fiero, naches, feeling respected), (2) other-praising emotions (admiration, elevation, gratitude, inspiration), (3) past-oriented emotions (forgiveness, nostalgia, relief), (4) future-oriented emotions (anticipatory enthusiasm, courage, determination, hope), (5) hazardous emotions (lust, schadenfreude, hubristic pride), (6) affectionate emotions (love, attachment love, tenderness, positive empathy), (7) arousal-defined emotions (euphoria, serenity), (8) violation-elicited emotions (amusement, awe, curiosity, positive surprise).

Graham, L. E., Thomson, A. L., Nakamura, J., Brandt, I. A., & Siegel, J. T. (2019). Finding a family: A categorization of enjoyable emotions. The Journal of Positive Psychology, 14(2), 206-229.

I made a note of this at the time because Tom Stafford tweeted:

"Want to be a scientist? It helps if your personality loads heavily on the ‘violation-elicited’ and ‘future-orientated’ emotions"

(I am feeling seen rn)

Forget all the political compasses and Myer-Briggs stuff. I want to know which of these 8 emotional families I chase more than other people.

4.

Yellow Drum Machine (Make: magazine) is a cute robot that drives around your house and plays the drums on whatever it bumps into.

Give yourself a treat and watch the video.

More posts tagged: filtered-for (124).