Near Future Podcast

Tom and Jonny sit down with inventor Matt Webb, the man behind Berg, Little Printer and the poetry clock Poem/1. They dig into how he built a practice around human-computer interaction, small groups and the physical world, unpack the retro charm of his ageing AI clock, and go behind the scenes of the Claude-powered agents running Poem/1's Kickstarter fulfilment.
Key takeaways
  1. Shape your role around your practice. Matt focuses on human-computer interaction, small groups and the physical world, rather than chasing a job title.
  2. AI is a rubber duck, not a co-writer. Its value is helping you think aloud without replacing your voice.
  3. Universal Basic Agency. AI can help non-experts cross the single technical hurdle blocking an idea.
  4. Agents can turn exceptions into policy. Poem/1's fulfilment agent proposes reusable rules for a shared SOP, so problems are solved once.
  5. Reports can reveal more than dashboards. Daily narratives surface patterns you did not know to track.
  6. Make deliberate time to learn new tools. Matt blocks at least a day a week because the gains compound.
  7. Building on a model means borrowed time. Poem/1 relies on a retiring model whose strange tone newer models cannot reproduce.
  8. Poem/1 is an invisible social network. All 700 clocks show the same poem at the same moment worldwide.
Chapters
  • 0:04 — Intro and Matt's career throughline
  • 5:56 — Leaving leadership to build
  • 7:28 — AI and unblocking imagination
  • 13:02 — Keeping AI out of your voice
  • 17:35 — ChatGPT and tldraw as a rubber duck
  • 24:12 — The imagination bottleneck
  • 29:04 — Universal Basic Agency
  • 34:38 — The story behind Poem/1
  • 44:44 — A retro AI product and invisible network
  • 52:08 — Agentic Kickstarter fulfilment
  • 1:02:54 — Reports versus dashboards
  • 1:06:40 — Quickfire questions
Mentioned in the episode
Matt
Companies and past roles
People
  • Jack Schulze — Berg co-founder
  • Matt Jones — Berg co-founder
  • Daniel Fogg — Matt's co-founder at Inanimate
  • Alex Hulme — founder and creative director of Approach, the industrial design agency that helped bring Poem/1 to market — approach.studio, Poem/1 case study at approach.studio/work/poem1
  • Tom Armitage — helped with Poem/1's firmware — tomarmitage.com
Tools and products discussed
Matt's own projects, mentioned
About the pod

Creators and Guests

Host
Jonny Burch
Co-founder of Near Future
Host
Tom Harman
Co-founder of Near Future

What is Near Future Podcast?

A regular podcast covering design and AI from the founders of Near Future, a boutique AI consultancy focused on teams that care about craft. We cover both what we're seeing on the ground and industry trends, ways of working and occasional guests from the design world.

Jonny: Hi everyone, and welcome
to another episode of the Near

Future podcast with Tom and Jonny.

This podcast is all about design and ai.

so I'm weirdly excited, to
introduce our next guest, Matt Webb.

when you think about design and you
of a certain age, Matt is someone

that you, will likely think of.

I remember being a graduate or there
or thereabouts and this weird company.

Berg was, doing all sorts of
interesting things, including making

a piece of hard work called Little
Printer that, was effectively a

receipt printer with a smiley face.

And it was like really early internet
of things where you could send it

your Twitter feed or it could send
you the print out, the weather for you

and you could put it in your pocket
and take it with you on the train.

I, Matt, I'm sure I'm, missing a hundred
other use cases, but it was just the

main, it was just beautiful, piece of
hardware and, design that lit everyone up.

I never got one.

great injustice.

I think my, my graduate salary didn't
quite stretch, maybe, I don't know.

But, since then, Matt, I mean, you've
done a hundred different things.

I've been following your blog,
a weekly written blog forever, I

think, I dunno when you started
writing it, but very, early on.

and it's an incredible read.

you've been running office hours and
actually the first time we met was

I booked in for one of your office
hours, half an hour on a Friday

in late 2023 or whenever it was.

And we talked about selling a business
versus pivoting and, various existential

crisis, crises I was having at the time.

and then since then you've gone into,
get into back into hardware, and I was

an early signup for Poem/1, which is
your clock, which writes a new haiku,

telling the time every minute, forever.

And so Tom and I are both owners of
Poem/1ss and we'll get into that as well.

And, and now you are running
Inanimate, which we will also get

into and is very much hardware
focused and, agent Agentic as well.

So I don't wanna give up away
too much, but it's probably clear

from the introduction that I'm
a huge fan and, overly gushing.

So I'll probably continue to do that.

Apologies to everyone
listening in advance.

if you feel a little bit sick in
the back of your mouth, I wanna

hand over to you, Matt, and end
this introduction by asking you.

When I reflected on everything I've seen
you do over the last however many years,

it feels like everything has logically
built on the previous thing that you did.

and it also feels like it's hard to put
you in a box, other than maybe inventor.

That was the closest word I got to.

So I'm just wondering if you could
start by just describing how Berg and

Little Printer got you here and what are
these through lines, through a couple

of decades of work that have, sort of
stuck with you and taught you more about

yourself and all of that kind of thing.

And we'll start there.

Existential.

Matt: Yeah, kind words.

Thank you so much.

wish I knew how to put it all in order.

So I'm a, I started in
as a software engineer.

co-founded a design studio with, Jack
Schulze and Matt Jones joined as well.

And then that was book.

And I suppose one of the things it
did was look for new opportunities

and technology and design to bring
new things into the world, perhaps.

So it was kind of like product invention.

And I think that's a, you know, that's,
that, that is something which I've kind

of taken through my entire career, right?

where, are the new opportunities?

What can you do something new?

And how that works is different.

It's very kind of context dependent.

So sometimes that means kind of
making, sometimes that means writing.

I spent some time as a consultant
editor and Google Research doing,

you know, editorial to kind of link
together their research in there.

Product.

Sometimes that means,
you know, helping teams.

So I was investing for a little bit other
people's money upset with RGA ventures.

But again, where are the new opportunities
and where do you bring them in?

And I think operating kind of very
early at that kind of like invention

stage makes it very hard to go.

oh, it's about engineering.

Oh, it's about design, or it's about
you know, strategy or go to market.

Because at that point everything
touches everything else.

So personally, I would say I know
enough about most things, badly,

like enough to be dangerous, but
it's the mix of all of them together,

which means it's possible to kind of
go, oh, we could do this with that.

There was a moment about, I don't
know, maybe Jonny, like near to when we

talked, where I was going through my own
personal, am I in the world, what do I do?

And I'd really lost my mojo with work.

And, I think then I took advice
from a friend who was just like,

really focus on your practise.

Like what do you like, do you do?

How do you do it?

And I started thinking about what
areas I wanted to work in rather

than kind of what role I wanted.

And practised for me really changed
a few years ago where I realised

that what I care about is I care
about human computer interaction.

care about small groups of people,
I care about the physical world.

And so I was just like, you know what?

I will apply my practise just
to investigating those things.

And I, changed the way I worked
to be much, much more hands-on.

And that was five or six years ago.

I've not looked back.

Best thing I did.

Jonny: Yeah, a lot of, maybe design ons
or whatever we're called now, would've,

you know, ended up in directing huge
teams of people and, you know, driving,

big change across multi-billion dollar
companies and going in that direction.

And it really, I think I noticed that
point in your career where you were like,

no, I just want to build shit basically.

I just want to like stay curious.

and

I think I've gone through a
similar experience basically where,

coming out of exiting a startup
and going, well, I haven't had a

design title for like eight years.

do I want to be a designer now or
do I want to do something else?

or do I want to be a founder mainly
or, and ending up trying not to worry

too much about that and start just
thinking about what do I actually enjoy.

and on acts not facts, which is
your consulting vehicle, website,

you have those, the Venn diagram
of those three things, don't you?

I think you say it quite openly, like,
these are the things I care about if

you want to talk about those things.

Great.

so was there a, some
nervousness around that?

Like what if I reject these obvious
career paths or this, this expected way

of doing things like judgement or like
lack of income or any of that stuff.

Was that part of the equation for you?

Matt: Think I was very lucky making that
decision about refocusing because I had,

I'd just come out of a period working
with a company where I was, you know,

doing products and design, but a lot
of it, it kind of started with coaching

designers and I was definitely kind of
on that track and that didn't work out.

You know, the company up and
I was like, what do I do?

And the chat GPT moment just happened
and I've been on the GPT-3 beta

some time and that space gave me the
real opportunity to kind of dig in.

And my sort of like, my meta observation
was, I remember like my first kind

of notes about GPT-3 was this thing
is capable of like creative ideas

or rather it's not capable, but
it's possible to use it together.

And that's new in computing.

And I looked there and I was
like, computing's not the same.

Jobs aren't gonna exist
in the same kind of way.

I don't know how I'm gonna
navigate my way through this.

I know is that we are
imagination bottlenecked.

We're not gonna sort this out by standing
in front of a whiteboard or a Figma.

only way you figure that out is
to get your hands dirty and to

see what the material tells you.

And so I was very lucky in that,
I had this kind of need to figure

it out and this new technology had
kind of arisen and that the only

way through it was to start makings.

So the first thing I made was, there's a
radio show I really love, BBC in our time.

I made the first podcast
trip back in the day.

So I like kind of deep affection
for it and I kind of wanted a

better way of navigating it.

And I decides to make a website
file and you know, this was like

right at the beginning, kind of, you
know, around the same time as chat

GPTI was like, oh, you know what?

I could make this using GPT-3, right?

Like use it for the data extraction.

Turns out I was one of the
first people to do that.

So it kind of got hacking used
and all those kind of things.

But was the moment where I was
just like, oh, this technology out

of its many different attributes.

unblocks me in of making work
in a way that used to kind of

require kind of a larger team.

And then I started carrying on making
more things like that and ended up

developing opinions about how we
work and collaborate with AI tools.

And that let me investigate the other
things in my then, right, like kind of

the physical space bit of multiplayer
then being able to kind of have a practise

of just like making, I had a rule for
myself, which is one gif every week, Like

investigate, communicate, share each week.

And after like a month of this,
somebody was like, Hey, can

you come and do that for us?

And that was a whole other opportunity
to start learning and building and things

just really kind of cascaded from there.

Jonny: A month of kind of
curiosity, pure curiosity, and then

like a green shoot of, hold on.

I could just do this.

maybe,

yeah, that's, really fascinating.

You mentioned, a gif every week,
you write a blog post every week.

How important has writing been to you as
a way to, I suppose, think through either

these problems or these directions you
want to head or even just the work itself?

Matt: Yeah, I think, I mean,
you'll kind of see, if you read

my blog, it's often very scrappy.

There are typos everywhere.

but one of the things I use it for
is to get my, like I start with kind

of something that I believe or a
hunch, and I try and unpack it and

explain why I think that's important.

process of linear, linear,

helps me put ideas in order and kind of
give them words and a bit of structure.

And happens then is I think about
the next thing, and that takes me

to an idea I didn't have before.

So for me, is not just a kind of a way,
or at least in my own blog of kind of

documenting or communicating, but it
is the process of having new ideas.

You know, lots of people
do that with sketchbooks.

Lots of people do it with their hands, you
know, other people do it in conversation.

it's writing.

benefit of writing like that is
there are no replies on my blog,

which means that nobody can kind
of go, oh, that was interesting.

And then just like bang
out a few words, right?

only way if so motivated they can kind
of like respond is to click on this

link called on Office Hours they have
to book a call with me, which means they

have to give up 30 minutes their time.

Right?

that means that the conversations,
I get through that by giving

up replies in the blog.

The conversations I get through
are incredibly high value.

They're so, you know, I've done
maybe 400, 450 calls like that.

We've done office hours over the years.

Talk with people from all over the
worlds, all kinds of backgrounds.

And they're always
incredible conversations.

And so that becomes part of
my process as well, right?

that's kind of part of having the feelers
out and understanding like kind of what

is going on with new ideas and culture.

Tom: Wow, that's a love,
love that approach.

have to ask the question, since we are
an AI and design podcast, is, does AI

feature in your writing process at all?

Have you found ways that, it can
elevate or challenge, or mirror anything

in how you work through your ideas?

Matt: feature one exception.

I'll tell you something.

So I tried once seeing, could
I involve AI in helping me

structure my thoughts into write?

And it came up with a turn of phrase that
was so perfect, I couldn't let it go.

And that terrified me because when
I, you know, like I said, like I

write in order to use the words that
come up to have more ideas, right?

Like those are Lego bricks.

And when I tried to lean on that
idea, that beautiful turn of

phrase, it wouldn't take my weight.

It was like I was
gaslighting myself, right?

It was like there was this idea that
sounded like it was from me, but it just

didn't, yeah, it couldn't bear any weight.

I couldn't use it for anything.

And it's scary, right?

To almost kind of have these things
in my notes that feel like, oh,

like my thoughts, but they're not.

So it well away.

so I don't let, I don't let
AI agents into my notes.

I'm terrified of that.

until very recently, I
wouldn't let them write emails.

Well, that's something which has changed.

And the one exception
there is, is transcription.

So one of the ways I love writing
blog post now is to go for a walk

and to talk into my watch 30 minutes
and then come back and turn that

transcription into an actual blog post.

I have a little workflow around that.

kind of address it to a sort of a, a
fake assistant, because then you can

kind of make outta band notes, right?

So I call her Diane's out
of, I was obsessed with Twin

Peaks when I was a kid, right?

So like Dale Cooper talking to Diane.

So I'll be like, you know, Diane
move that idea like to the previous

section so you can kind of have this
kind of like conversation, which is

both transcription and outta band
instructions about editing simultaneously.

And then I'll take those words
and then turn it to a blog post.

And those are, that's a
really fun way of writing.

But apart from that, I'm like,
you know, won't live it in it's.

Jonny: That's more the analogy of the,

Tom: Great.

Jonny: assistant with a, a notepad
while you dictate into your,

into the air as you're wandering around
your office, and then they come back

for, and say, this is what you wrote,
and then you can sort of review that.

I found very similar writing, which is,

it's almost too good.

Like I'm not as good at
writing as it in some way.

Like, it doesn't feel authentic.

so yeah, I completely agree and I, dunno
how much, product design or sort of,

visual design or that kind of stuff that
you're doing at the moment, but have

you pushed into anything else with ai?

Have you found AI to, be similar in
other aspects of your creativity as well?

Do you do not trust it to sort
of complete something that is, I

suppose more on the design side
rather than the, the writing side.

Matt: I use, we use AI a lot, right?

So, bit of context, you know, early
stage startup called Inanimate.

I'm co-founder, co-founder Daniel.

we have a kind of a small team of
partners, freelancers, employee, and

we're putting AI into the real world, and
we're doing that via physical devices.

So we're making hardware, we're
gonna be bringing out some stuff

towards the, you know, towards the
end of the year, which are some very

early prototypes to kind of get our
interactions and our technology out

there in the world onto people's desks.

just short run, you know,
not even products, right?

And, you know, that's, that, that's
a lot of different streams to be

running at the, the same time.

And there's, you know, there's
a limited amount of time,

course, Claude code, right?

Like tonnes and tonnes of that.

That's fine.

You know, taken for granted.

It's been wonderful watching our embedded
engineer, using an AI plugin he developed

with PCB software to review PCB layout.

That's kind of a use useful tool.

interesting is that one of the parts
of my process is that, about two thirds

of the way through design work, will,
just like make the landing page for the

projects it's a really interesting filter.

You've kind of, you've kind of understood
what the thing is, you've been making

it, you know, but there's a bit where
you have to kind of go, is this gonna

tell the story to other people, right?

Like, not just kind of what the design is,
but how it's kind of understood physical

things with kind of software behaviours
and making that landing page is.

It's not gonna be the landing page, right?

So it's kind of redundant work to make
it good, but there is something useful

in kind of like helping to think out
the information hierarchy the messaging

hierarchy and how to tell the story
and which bits you're gonna highlight,

where it's useful to go through the
process because it what it tells you.

And I tried for the first time the
other day to go, well, what if I could

do this via transcription, right?

So I got a coffee, went and sat in the
park, I talked about this collection

for putting together and just everything
on my mind, like 40 minutes and came

back and with, GPT in that case to
make a microsite to explain these

using, do you know, tldraw the,

Jonny: Yep.

Matt: whiteboarding app, right?

They have, tldraw Offline, which is a
version which has of agents skills that

agents are really good apps, like seeing
it and drawing on it and iterating.

And I had it make me the microsite based
on my transcription in tldraw offline.

Then I covered it in post-it notes
and got it to read the post-it

notes and do another iteration.

And it was useful I think, not because
this is anything which is gonna go

in front of people, but because it,
it helped organise my own thoughts

and gave me something to bounce off.

And that's been a kind of
an interesting application.

Jonny: Like AI as editor
rather than writer.

is that reductive?

Matt: AI is rubber duck.

Jonny: Yeah.

Matt: rubber duck phrase?

Right.

Jonny: Yeah.

Matt: know, the, kind of, you know,
you're trying to figure out a difficult

bug, so you need someone just to talk to.

I think there's a little
bit about that, right?

Jonny: Yeah,

Matt: Yeah.

Jonny: if you were rambling in
the park to yourself then that

would be probably ably insane.

Whereas if you're rambling at least
to something that might have an

opinion about it, or, I mean maybe
if you had an actual rubber duck in

the park, then would that be useful?

I dunno, but.

Matt: I do.

I do wonder what people think when
they see me walking around and

talking, because like at a certain
point you're like, that's fine.

They're on a phone call and they've
got AirPods on, and if they're

sitting opposite, they'd be like,
that guy's been talking nonstop.

Like 20 minutes.

Like, let the other
person have a bit there.

Come on, what is

Jonny: This.

Matt: No, it's great.

And there is something, there's something
great about like, you know, I was

debugging some firmware recently and I had
the piece of hardware plugged into my Mac

mini at home and I was the debugging just
on my phone, just like typing in, like try

these things out and walking in the sun.

there is something wonderful about kind
of like walking in the sun, debugging

firmware, walking in the sun, thinking
about like how a site is gonna work and

how the messaging hierarchy works and
how that impacts the product design,

which is, a different way of working.

I think it provides a looseness,
which I think is really important.

Tom: I think so much of what you're saying
feels like it's built around this tight

feedback loop of making something, writing
something, getting input from others, sort

of reducing the friction and how you that
into the world and iterate on it quickly.

I'm curious in relation to that,
and I'm curious about your energy

throughout this whole experience.

I feel like there's been a big change
in, for a lot of us, working with AI

in the last few years of things feeling
increasingly tiring or exhausting

because you're expending so much
more energy and a lot of what you're

just describing there sounds like
very, mentally intensive activities.

Like the thing that used to be the
break to go to the park is now a 40

minute walking, production session.

And then presumably you get back to your
laptop afterwards and, take that and

start moulding it into something else.

Like, is there any, are there any changes
to how you balance or how you create

space for letting your brain absorb
what is happening without constantly

making, or is that part of your,
personality to just producing and then

it sort of works itself out in the wash?

Matt: that, you know, you sort
of get that a lot with coding

agents especially, right?

And I wonder how pilots deal with this.

Well, you know, airline pilots, you
know, fly by wire, you know, all

the rest takes over and they only
get involved in difficult bits.

But at least if you're flying a
plane, they don't kind of say, well

that boring kind of eight hours in
the middle, you've now got to go and

do the difficult landings elsewhere.

is an element when you're working with
half a dozen agents at once, you are only

involved in the really intellectually
difficult bits as opposed to the kind

of the very pleasant sort of like, you
know, doing the knitting in the middle.

It's exhilarating, isn't it?

this whole world.

And you definitely need to there are
moments where you just kind of take a

step back and just go, I just need to
like think about this, build my own

mental models, relax a bit with it.

but you know, it's also a startup world.

This is not the time to be kind
of, you know, pitching back.

I've got some, you know, we've got some
very kind of definite opinions about

what we're building and it's a matter
of like making it work at the moment.

Jonny: we should switch onto,
Poem/1 and inanimate in more depth.

But before we do that, just to
sort of round out this, this first

section, you have carved out space
for yourself to explore in this way.

Tom and I both actually have over
the last year, part of what we're,

we've been doing intentionally is
allowing ourselves to play with this

technology and get a feel for it.

A lot of, people working full time,
you know, as a designer inside an

organisation, which sets the parameters
around what they can do a little

bit more, don't necessarily have as
much of the like free thought, you

know, zero to one type work to do.

Do you have, I suppose, do you have any,
thoughts or advice as maybe, but just

around what everyone should be picking
up right now, if they're not already

to just help inform their worldview
around what's coming or like what,

how they should change their practise?

Matt: I think there's, I would say
companies are, I've used this phrase

before, imagination, bottleneck, right?

So, you know, for the last like
kind of 10, 20 years we've been

working with digital and we all
kind of know what that looks like.

We know what good looks like, we
know what can be done with a team

of six in three months, right?

Like everything has got a kind of a,
a shape to it that we understand and

we figure out the best thing to do by,
know, doing wide frames and standing in

front of sopping, post-its, that isn't
the case with these new technologies.

We don't know the best ways to
design them and, we have to be

really sensitive to kind of what
people feel when they're using them.

and these are often new
feelings that people having.

You know, Tom, you mentioned
one about the, overwhelm, right?

And there are other kind of new
feelings that like users have as well.

And the only way to figure
those out is to make things.

So, you know, where this kind
of reminds me of the very early

days of smartphones taking off.

It's like we do not know what is
possible, what's easy, what's difficult,

you know, our intuitions are wrong.

that's at a corporate level like
that, that has to happen somehow.

Any leaders out there like make time
for that sort of like build and share.

But on the personal practise,
of our investors is Betaworks.

And earlier this year we were with
Betaworks in New York with the

rest of the cohort of, founders.

Another, you know, another
bunch of, startup founders in

our kind of camp they call it.

And it was inspiring to see how
those people worked because they

were embracing collaborating with
agents, like at a level I hadn't seen.

And I realised at that point that
I need to invest more in my own

tools and my own working practises.

And I kind of, I would say that on
average I would probably invest,

I dunno what, like a day, a month,
less into kind of, oh, you know,

try a new note taking app, right?

You know, that would be it.

And I consciously said, this needs
to be about a day, a week or more.

I need to be going, how do I, you know,
how do I use like, well, know, I don't

know, like my desktop to drag my browser
to like, know, make the kanbans for

me to, to run an autonomous fashion.

Like I need to learn these things
and to be really kind of like about

finding points where I can use
more automation and intelligence.

And I'm not gonna find that
by just hoping it comes up.

It needs to be a deliberate effort.

And that was, you know, I, made
that kind of conscious shift and

that was like really valuable.

So I would say just like, it's not about
kind of exactly what tools to use, but

just increase the time you give yourself
to figure out what to do because that is

an investment that will compound really
quickly over a period of like months.

Tom: True do.

I'm curious, do you have any hunches
on whether that is a reflection

of this current moment with AI
or that is a practise we will

continue to be doing years from now?

Matt: Think it's a difference today
based on how it was yesterday.

I wonder what it will be like tomorrow.

I dunno whether the exponential
is gonna, like, flatten off a

bit or curve up or stay the same.

Right.

I think I am a, it's really difficult kind
of, I, like to say like, I'm a believer

in kind of like investing in tools and,
you know, this is the man who's had the

same way of writing for the last 20 years.

So, you know, or you know, five or six.

Anyway, it's very easy to spend time
going, gonna think about new processes

and just like swap from one piece
of notetaking software to another.

So there's got to be some kind of
discipline at least to go like,

how am I gonna do this differently?

But I, wonder, that's
an interesting question.

I wonder what it would be
like in another two years.

Tom: I think a lot of these conversations
have that, that question has been

circling around like what is a
specific thing as a result of the

state of the technology right now,
folks from larger organisations.

There's more of a question of how much
is of this is a temporary infrastructure

to introduce now and then a year from now
that will just be established and the team

will be set up to work in this new way.

And we're just sort of on the edge
of this transition is happening.

But I think it, hearing how you work
and your own practise, it feels like

it is fundamental to how you are, you
will continue to iterate in that space

regardless of where the technology is.

It, sounds like you, that sort of
feedback loop that is so central to

how you work is critical doing this.

So I can imagine you being who would
invest more in that space than others.

Matt: always more to learn.

I think one of the ways I think about it
is that these AI tools provide what I'd

call like universal basic agency, right?

There are a bunch of projects that I've
not done before because there are certain

pieces to them that I'm not capable of
doing and that blocks the entire project,

and AI just gets me over the hump with
something that might be quite small that

somebody else would find very, easy.

But if I wanted to do that project, it
would be a question of enrolling someone

else In this kind of picture in my
head, I'll give you an example, right?

So for years I had this idea that
I wanted an arrow on my phone that

would points to Centre of the Galaxy.

Ridiculous idea.

About 10, 15 years ago, I used to be
able to do this in my head, right?

You could kind of like visualise
it, the earth has got kind of an

axial tilt, goes around the sun.

You can think about what time
of year the solar system itself

has tilted, and then there's the
kind of the centre of the galaxy.

And so the earth's rotating,
the earth's go around the sun.

It's all these different
tilts, time of day.

And you know, the central black hole
is 26,000 light years away straight

through the constellation Sagittarius,
Why do you need to know that?

Don't know.

That's what I thought about at the bus
stop when I was waiting in is LinkedIn.

over the years I forgot how to do this.

And I was like, you know what?

It'd be really good to have
a phone app to do this.

I spent like two years trying to persuade
people to like, help me build that.

I couldn't build phone apps, right?

I can draw an arrow and find of say
how it wants to behave, couldn't do it.

And that was, I talked to a friend and
he was like, yeah, I'll totally like, you

know, I'll totally help you build this
app and, you know, never got round to it.

And then one day I was like, oh, this
chat GPT might be able to do this, right?

So I spent a day kind of like
copying and pasting code between

the chat GPT window and Excode.

was sort of previ coding.

And it got me over the hump.

There was a problem about how to
combine rotations, which requires a

mathematical technique called quaternions.

had to learn that as well.

but it got me over the hump of being
able to, you know, I can't make apps,

Jonny: foot.

Matt: but then it worked, right?

And so then I released this app and I'm
like, this is like just for me, right?

Absolutely blew up.

And I started receiving emails from
people who are like, oh, you know what?

I'm gonna, a really
difficult place in my life.

I sit outside in the middle of the night
and I smoke a cigarette, and I think

about my father and I look at your app
and it, you know, it suits me to know

that there is one fixed point that we've,
you know, we can regard as like steady.

I'm just, you know, imagine waking
up in the morning with an email

like that in your inbox and it turns
out that all of that was worth it.

Would, I have been able to persuade
anyone to make this very, very speculative

thing with me, for me to put their time?

Probably not.

So what happens is like your, as you
know, your capabilities grow, your

kind of desire to make more grows and
you can make more things which are

based on even more esoteric hunches.

I mean, you know, Poem/1, right?

Like case in point, there's no way I
could have done that my end before.

And so I think, yeah, like
just on learning more tools.

it will be interesting to see where
this goes, but I, the thing I'm excited

about is the individuals and designers
and the artists who can now put their

kind of pre-verbal imaginations into
the world in a way they couldn't before.

Tom: And, excellent segues Poem/1 as well.

I wanted off this, this next segment with
a, a little story, which sets the scene

for how this conversation came about.

we were at a dinner recently.

I was lucky to sit next to you.

We were chatting.

I think that was a good conversation
around what is boring and sort of trying

to define the state of boredom and how to
achieve that great dinner conversation.

Much more interesting than it sounds.

but at dinner I also brought along
my Poem/1, my, physical device with

the, e ink display, which, shows you
a new poem every minute for eternity.

And you, held this device.

I was a little bit nervous about
showing you, I thought I would come

across as either a fanboy or just
some weird person from the internet

who's brought this physical device.

wasn't sure what kinda reaction
it was gonna get on the

fence about even doing it.

And then you were studying this,
this device looking closely at

it, felt like for, a long time.

And then, eventually we had the discussion
about the, the screen not being at the

level of quality that you were hoping for.

and, the, the original device was
great, in my opinion, you have sent me a

replacement and the new is even better.

But I think there's something about
this, that, that set the scene for a

conversation around how to use AI with
the aspects of Poem/1 that help you the

things that you are either have limited
capacity for or you don't enjoy doing.

And that was sort of the seed of
the conversation that I thought

would be so interesting to have you
on talk about through your work.

But perhaps preface that, before
we get into that, maybe you could

just give us a bit of an overview
of Poem/1 as a project for you.

what was the impetus for it?

You kind of set the scene a little bit
with ChatGPT and, sort of seeing its

creative potential, but I'm curious what
went from there to actually having this

idea for a device and putting it out into
the internet and getting lots of people

to support that, help us fill that gap.

Matt: kind of a ridiculous
story of, many twists and turns.

I'll give the quick

but I want to say something about
the prehistory of the AI clock it

was called then, it came from what I
would call the reverse Turing test.

So I was building a prototype with
a client where this is very, early

again, like GPT-3 days, right?

G PT 3.5.

and the user needs to be able
to speak via WhatsApp to an ai.

And I made something just to kind of
connect these things up, like what the

prompts are, like, what it was about
doesn't really matter, it's very important

in any project to get to end as quick
as possible so you can feel what it's

like to have something which is not
human text appearing in your pocket.

Right?

that's unusual.

That's gonna feel different.

So you're like, what's that feel like?

And made this p.

You know, something comes back
on the phone and I'm like,

I don't even, I built this.

I don't believe that's ai, right?

Like at, 'cause at the time you
would default to being human text.

I'm like, I'm gonna demo this.

And they're gonna think either
it's automated or they are

going to, type something in.

It's gonna come back and they're
gonna go, Matt is just there

at the other end just typing.

Right.

And I've experienced this before
where I was like, I was evaluating

a company which could, they claimed
they could send back iMessages.

was like, how do I know it's not
just like the founder doing that kind

of YC thing, like things that don't
scam, like literally typing things in.

I asked it to say the alphabet backwards
and it came back really quick and I'm

like, oh, okay, that's a computer.

So for this prototype, I was like,
how do I prove this is actually an ai?

So I was like, I know what I'll do.

It's like proof of life, like, you
know, the kind of people that hold up

the newspaper to go, this is happening.

I'll get us to do
something about the time.

And, the thing that I can't do really
quickly is make a poem about that.

So I'll make a poem about the
time the user type something in.

So they would go, hello?

It would send back a poem about the time.

So that was the thing
I did with the client.

I came downstairs one morning, I have
an e ink clock on my shelf that, is an

open source project from the Connected
Environments programme at, UCL.

Been there for a few years and I
was like, what if I put that poem

on stack thing on that clotter?

So I did that and it was weird because
this is, you know, we don't really

have a culture I think of like changing
texts in the world in English, right?

Or even text as decoration,
like live, laugh, love, right?

Like that's kind of maybe it.

Right.

And this text was strange because
it was really it was kind of, kind

of a get up and go kind of vibe.

it would lie about the
time a, a rhyme work.

there was this kind of like
aspirational nature to it, and

I was like, this is strange.

So I sort of posted 'em on
Twitter, completely blew up.

Everyone started retweeting it,
ended up very quickly in New York

Times as an example of like, kind
of what people are building with ai.

And I'm like, you know what?

I want to get this into
more people's hands.

So, I started gathering people's
email addresses, and at the time

I was like, I'm gonna make a kit.

I was like, I tried to make a
kit, and I was like, couldn't

guarantee that people get a piece.

It's like, okay, I will try and give it
away to someone, and all I want in return

is one of these clocks to put on my shelf.

And so I talked to a retail company
that would do some product development

a manufacturer got in touch as well
because they'd seen this kind of thread

they'd seen the newspaper and they
were like, okay, we'll, do it right.

And then for whatever reason, they had a
change of heart, like right at the moment

where we were going to go into, like kind
of the industrial design development.

There were mockups and everything
and, that fell apart just like wery.

They changed their mind.

at that point I'm a bit like, come on.

I was so close to having
this thing on my shelf.

Right.

And I remember looking back at my, I made
some kind of notes about how to bring

it into the world you know, option A is
like a kit, option B is like, somebody

else makes it, option C is Kickstarter.

And I'd like, I'd crossed it out
and I was like, never do that.

This is a bad idea.

And I then ran into a friend Alex from
who's, founded Approach, which is an

industrial designs agency, he was like,
you know what, we can help with that.

And it gave them something, you
know, it was great to work with

'em because they really, like, they
work with the contract manufacturer.

I know.

Well, know much better now, but new
of them, they got to show off some of

their thinking about AI and we worked
really closely, on the subjects.

I wrote to another friend, Tom Armitage,
to help out on the firmware as well.

together on that.

And then it, so it's been,
you know, a bit of a kind of a

team there to make it happen.

then yes, as you know, eventually
I was just like, I wanna

see this thing in the world.

Take it to kick start.

I thought that was gonna
be a very quick process.

It's just an ePaper display in a box.

what could possibly be
difficult about that

years past?

You eventually have it on your shelf.

that was my evenings and weekends,
for a good period of time.

And, you know, as you mentioned
about the screen, one of the things

I really obsessed about was the
Matt finish on this ePaper screen.

Getting that right, the QC to kind
of get that happening, really tough.

I've got it here.

And, but that, and that finish matters.

And when I saw yours at dinner, I'm
just like, something is not right here.

didn't come all this way to have
somebody with a clock where they've

forgotten to put the screen film
on and, needs to make good on that.

Jonny: Amazing.

I,

Tom: Amazing.

Jonny: I got my Poem/1, I think I posted
about this, but even though I knew exactly

how it was gonna work, and I'd seen all of
your updates and all that kind of thing,

I held it in my hand and, you know, it,
it took two, two years, I dunno well over

a year to get from me supporting that
Kickstarter to me holding the clock in

my hand and all of the, the pain that
you went through, I saw all of that.

but what also had happened is the
technology underpinning it had changed

so dramatically that what had maybe
been extraordinary when you shared it on

Twitter when you first built this thing.

the, the technology itself
isn't extraordinary.

In fact, it sort of feels like,

I don't know if retro is unfair,
but like, it, sort of feels

lo-fi compared to everything else
you can do with this stuff now.

So it had a, a different impact
on me than I thought it would when

I placed my Kickstarter order.

and as such, it sort of feels
more meaningful, it feel,

feels like a time capsule.

and it, because it doesn't have to
be a demonstration of extraordinary

technology, it can just be what it is,
which is a beautiful thing to, look at

and a hilarious and sometimes overly s
mushy like rhyming coupler that appears

on your, on your, device every minute.

So yeah, I, kudos 'cause tho those
people didn't order a clock and

haven't been following along.

Won't have seen what it took to get there.

But it is a beautiful, I've just
moved house, so I dunno where mine is.

Otherwise I'll be
brandishing mine as well.

But, it's in a box still, I think.

But,

Matt: Retro, retro is the right word.

I mean, very, strange to have made like
the world's first retro AI product, right?

it's AI from a more innocent time,
because, you know what, it do?

It shows a rhyming couplet
for every minute a day.

That's 500,000 poems a year, right?

That is, you know, the highest poetry
velocity of any consumer appliance, right?

Like, I dare you to find, one else.

But at the same time, it was far
enough into thinking about AI that,

the clocks show the same, poem time in
the day, same time for that day, right?

It's different.

Explain.

and that's because I, you know, like even
at that point I'm like, I'm not sure I can

justify data centres spinning up with the
carbon footprint and the water for like,

you know, something this frivolous single
person and weirdly, that had a kind of a,

it does provide a sort of an extra layer.

The fact that, you know, are 700 clocks
in the world, which we're all kind of

showing the same permit at the same time.

Is that when you look at it, somebody
else might be looking at the same thing.

And there's a kind of a, somebody
called it an invisible social

network there, which I really love.

There is this kind of like,
possibility of communion someone else.

And it, yeah.

It's, you know, one of the, one of the
other problems about it being older is

there's a whole kind of poetry machine.

I mean, I'm the poet here, right?

I kind of live with this on my,
kitchen for, you know, a year and was

like tuning the, machine behind this.

but it's built around the model,
which is very soon to be turned off.

And every time I try and update
the machine, so like, move

away from that model, the newer
models are just too domesticated.

I can't get the variety in.

Jonny: Four Zero.

No.

Matt: real kind of challenge there.

yes.

Jonny: Yeah, famously, famously romantic.

Matt: Yeah, it's really tricky.

So yeah, this is, now the, active
sort of like, what do I do?

What do I do about that?

Right.

You know, having gone through, you
know, the design and the manufacturing

and the sourcing and then fulfilment,
and that's where the agents come in,

dealing with FedEx and all the rest.

Right now I'm onto, well, how do
I, like, how do I keep this live?

I'm, you know, I've got some
more in the warehouse on Shopify.

All those kind of things.

people will still be able to get their
hands on them, but the, the poetry is the

thing that kind of really occupies me.

And that was its own
journey as well, right?

Like kind of to kind of make something
which was frivolous to begin with

then to kind of really go through
like, you know, who is, the poet here?

What is my role?

The machine is not the poet.

Right?

Like that, you know, I kind of
say like that, you know, chap,

GPT, composers, poems, or clock.

Like, it's, me.

I have, I stand, you know, I have to stand
behind this it has to be intentional.

And so that's, you know, that,
that was unexpected as well.

Jonny: fascinating.

Matt, I'd love to learn about what
Poem/1 then took you onto, and it

feels like there is some connective
tissue between that and inanimate.

but, in your de describe the, the
journey of this like art project,

this, as you say, frivolous, you
know, beautiful, but probably not,

infinitely scalable thing that you built.

and how you sort of either, either
parlayed that or, moved that into

inanimate and what is inanimate.

Matt: I mean, one of the wild things about
doing poem is like realising what can be

done with a small group of people now.

You know, last time I did, know,
manufacturing with Burger, it was

like that studio was full of people
from many different disciplines.

And while we were making little Printer,
I think there must have been like

kind of 15, 30 people kind of in the
studio and the kind of the core team.

Like so lucky to be working
with so many incredible people.

And the idea of doing something which
was, you know, like, like I said,

there are kind of like people I worked
with that kind of like more on my own.

It's, kind of ridiculous, right?

And part of that is like the China
ecosystem, the Shenzhen ecosystem

and the UK system ecosystem are like
way more developed than they were.

to kind of work together like that.

some of it is just kind of
making s smart decisions about,

putting, you know, how what the supply
chain is behind different things

like that, that's more sophisticated.

and I think one of the, you know, aspects
of doing Poem/1 was it just kind of said

to me, it is now possible to think in
credibly about hardware in the world

from a kind of a startup perspective.

And maybe kind of, it hasn't
been from the last like 10 years.

I think the other thing it made me
really feel is like, it is magical

to have something which is kind of
like which isn't kind of inside an

app window in the world around you.

How does that work?

And since the beginning of doing poem,
I really started asking myself like,

how do you have AI in the world in a way
where, you know, regular people use it.

And there are all kinds of like
really interesting issues when you

start digging against that Right.

You know, interaction.

How are you gonna interact
with, you know, in the world?

How do you have really low later,
you know, kind of inside the 150

milliseconds, kind of like cognitive
loop of kind of instant in.

How do you do that when inference is
at least a kind of a network half away?

how do different things
kinda work together?

how do you generate more interfaces?

And it was on those kind
of questions where I.

I started developing some technology
and some opinions about the

interactions would start answering that.

And I think this would've gone nowhere
except for an old trends Daniel, came

over for dinner because he was kind of
thinking about kind of similar issues.

we started talking about like, okay,
but what does this mean in the world?

How would we actually do this?

Like, how would that come into the world?

And outta those
conversations came Inanimate.

We were like, we think we
can actually do this, right?

we can turn the room around us
into being a, ui, which increases

the agency of people in it.

but the way we'll start is with
objects that people understand as

products that they kind of allow,
you know, can be understood and

get invited into people's lives.

And we'll start in a quite
humble way and we'll take it, you

know, sort of like step by step.

And was the, origin of Inanimate.

And we saw, you know, raised
a bit of exploration money and

been the story of the last year.

But equal parts design, technology,
you know, product design, understanding

how people understand things and
and kind of knowing what is now

possible with the Shenzen ecosystem.

Jonny: So what can I just
ask about exploration money

that has piqued my curiosity.

as much as you can share, like what was
your story that you told the investors

of the angels or institutions or
whoever helped with exploration money?

Like what was the, were you
able to tell a story of a grand

vision and an enormous business?

Or was this actually just, there's
something interesting here and,

we would love to, explore it.

Like what, was the, how
did you go about that?

Matt: so we were, we pretty much did
this kind of digging our own until

there were a few technology prototypes.

when we start using some of that, and one
of the core bits here is a of software

that we open source called Resident, which
lets you sort of hot reload, new behaviour

into products kind of on the flight.

And between that a couple of other
explorations, we were like, all

products are going to work this way.

Like, it just became clear like
you start using these things, it's

Tom: It.

Matt: this is how things
are going to work.

And at that point, you know, where if you
meet the right investors who are prepared

to kind of go, not only, you know what
this kind of feels like, you know, you,

might be on something here, but also you
are the, people who could pull this off.

That was what, that was how
the pre-seed came about.

And we had to have, we have a great
group of investors involved in that.

so it was very, early prototypes,
but enough to kind of go, yeah, okay.

There are some kind of core assumptions
here which are kind of born out.

and

Jonny: But

Matt: go.

Yeah.

Jonny: I'm a, early, I'm a, I've
just started exploring resident.

I have my little M5Stick here, which
I dunno if, those people that are

watching can say it makes a, it's
meant to say hello Matt, but this,

this hardware doesn't, have a real
speaker, so it just beeps at you.

but yeah,

Matt: It's

Jonny: excited to keep exploring.

Tom: It shows on the

Jonny: Yeah.

it does say it on the screen.

Yes.

okay.

Well I realised we skipped over, one
topic and we will cover that last

before we get to quickfire questions.

So, Tom, I'll hand over to you for.

Tom: Yeah.

A gentech fulfilment.

So Matt, you, well there was so
many different directions we could

take this conversation in and Poem/1
led Oh, so many doors to open.

But that dinner conversation the other
night around, specifically around how

you use Claude agents to help with
the fulfilment process around Poem/1.

if you could share us a, a bit about
that wonderful world and, how has that

experience been fulfilling Kickstarter
orders in this new era of AgTech help.

Matt: fascinating things about, you
know, there, there's this kind of

line like, you know, they call it
hardware 'cause it's hard, right?

And weirdly, I think that like
connected hardware is not,

you know, it's not difficult.

you know, I meet engineers who are
dealing with things like the sort

scale of the internet or the kind of
the, depth of computer science that

like, I've got no way of kind of
like, you know, even comprehending.

But the thing with hardware
or with product is there's

just lots of different bits.

It's very, broad.

And one of those very, very broad
pieces is when you've got your products

in the world, you've got to get 'em
into people's house, fulfil, right?

You know, the factory has
made a bunch of things.

They've put 'em on pallets,
they've gone into a warehouse,

they've got to arrive somewhere.

That to me, is as fascinating as
the ID as the word wrapping the

firmer or the interactions around
what happens when we hit the depart.

And I'd kind of everything else in
the process, I touched a little bit,

but I hadn't thought about fulfilment.

I was just like, that is
a piece of the process.

I will not think about it.

Turns out is a really good
ecosystem for this, right?

So my, my manufacturing is in
Dongguan, which is a factory, about

an hour's drive to North Shenzhen.

over the water from Shenzhen is Hong Kong.

Hong Kong's a really good place to
ship from it's a, you know, it's

a, you know, very open borders.

You can get there from anywhere in the
world, including from mainland China.

and it turns out there are warehouses
which specialised in crowd funding,

fulfilment and like, you know,
but later for the Shopify, you

know, that kind of shipping.

So I was like, you know,
jobs are good one, right?

Like kind of put the
options in there back a kit.

Everyone kind of gives their details
and I'll just kind of upload a big kind

of CFD and Floship and send everything.

Should be easy.

Was not easy.

The first hundred clocks, which, you
know, they're kind of the early test,

they went out fine about half of the uk,
half of the us, some scattering elsewhere.

Even though I made people promise I would
not shit for anywhere else in the world.

People moved in the meantime,
fine, all went out fine.

So then I ship out another 600, wake up
one morning bank holiday weekend, wake up

one morning I've got like 15 messages from
my warehouse who are like, know, has got a

question about this package that went out.

can you just tell us what it is?

And I'm a wrote couple of lines.

And then one of them linked through to a
page on FedEx's website where there was

this kind of like big long form fill in.

It's like, this is weird.

And I sort of grind my way through that.

It's like the family was out for
the weekend, I'm like, sunny, right?

So I just go and grind
through those things.

Just I'm finishing like another
10 or 15 or it's like, okay,

this is now getting tricky.

So I think, okay, I'm gonna start,
write a little called skill for this.

Drag my browser.

I've never done that before.

You know, been a bit cautious
about these kind of things.

and I'll give it all the
background information.

I'll write a little Claude skill where
it can, drive my browser and just kind of

like fill in those things for me, right?

So what I'll do is I'll
just give it the URL.

I'll go there, I'll wake up next
morning, another hundred I queries

from FedEx and the warehouse of
the right, all slightly different.

I'm like, okay, that's a problem.

I can't deal with this.

Now.

I am not the kind of person that likes
AI agents in my notes or in my email.

find it like psychologically
destabilizer, but at the same time,

I do not have time to respond to a
hundred different requests for FedEx.

Especially filling in
this very dedicated PDF.

They gave me where they were like,
it's just, you know, a declaration

form about a clock where all the
questions are we think it's a watch.

Is it a watch?

No, it's a watch.

How big is the movement?

Ah, sounds like the movement.

The movement is the, bit of
the thing that makes it go that

you look right, not the case.

that's really small boxes.

PCP.

Right?

It's really small.

That means it must be a watch.

Are we right?

It's like not a watch, it's a clock.

How many jewels does it have?

It doesn't have any jewels.

Right.

And so they, I think they
thought I was smuggling watches.

And so I built, ended up building a
whole bunch of board skills to like

deal, like read my email, bring out
all these things, and then people

start getting the clocks and they start
asking me questions, support questions,

forgotten about all of that as well.

Right.

And you think there's a few of them, but
there's a kind of a scale problem, which

is just like people have very innocuous
questions or something doesn't arrive.

or like perhaps something is broken
or the firmware's not quite right.

There was a new networking problem
that arose between the test units and

the new ones just be tracked down.

So I'm tracking all of these.

I'm like, I can't do this.

MIT train plot can help.

Again, the problem with agents is
you give 'em a problem, they grind

every single problem into dust, right?

They don't know where the
bottom is, and they try and work

everything up from first principles.

At least this is my experience,
and they can work a bit

differently in different cases.

So want that, know, I don't want it
to kind of like accidentally miss

an email one time, reply to somebody
twice, you know, decide one answer

for one person networking problem.

Another one for somebody who's
maybe got kind of like the screen

finishes and what they wanted.

I need policies for that.

So what I started doing started thinking,
what does my agent need to work?

let me, actually just show you, right?

This is my ops screen.

So this is a website which is not meant
for humans, it's just a picture for me,

to, see what the agent is seeing.

So my Claude communicates with
the backend website via an MCP.

says MCP served for this.

That's the way it interacts.

This is a nont interactive website,
this is basically like a cache of

all the data, coming in from Floship.

My warehouse BackerKit, which kind
of manages, you know, backer surveys

and collecting fulfilment reads.

My email caches it there.

And then it triages issues
which are appearing in email.

it has a kind of a basic CRM,
which is basically a linking table.

The idea is that like there's very,
little beta kept here, but it's here

to maintain state for the agent.

And I'm not gonna go into any
of these because I don't wanna

kind of reveal kind of PII.

but Tom, we've got this one right, which
is, this is a very, ugly screen, right?

But it's not meant for me.

This is a log of my, like
pulling out from my email.

these are kind of pull them from Floship
so I can see kind of what the orders are

and I can kind of track 'em through here.

there's a kind of an LLM written kind of
description about like what the, going on.

And I can kind of run this every
couple of days and triage, new issues.

I also have, you know, I have a kind
of a stats page over here where I

can see where clots are being turned
on, you know, if people are upgrading

their firmware and things like that.

this is interesting, I think because it's
the first time I've built something where

the important bit is I'm building new
ground troupe, and then the agent itself

is figuring out how to operate on that.

But it's not having to figure out
things through first principles.

So the agent skill started
as, read this via the MCP.

see if you can identify issues.

If you have a way of dealing with it
in a document called the SOP Stand

Operating procedures, then do that.

Just apply that.

Ask me to confirm before
you send anything.

If you don't, then propose what our
policy should be, approve it, add

it to the SOP, and then apply it.

And over time the SOP has grown.

So there are now a number of very
detailed policies about how to

deal with different issues, which
don't have to be figured out again.

And it's that kind of like compounding

Jonny: Self-healing.

Yeah.

Matt: Yeah, I couldn't have
run this stage of the process

without Claude doing this for me.

I also couldn't have done it had I
just let the agent loose on my email

and loose on the warehouse APIs.

there is something about, and I think
this is a kind of an interesting pattern

to kind of go, I mean I think about
this sometimes as like context plumbing,

but it's structured context, plumbing.

How do you make sure things flow to
the same place so that doesn't become

a new source of truth, but is just
pulling together source of truth from

elsewhere and then have the agent
operate on that, but in a way where

it has kind of compounding processes.

And that's something I'll be
carrying into kind of other projects.

Definitely.

Jonny: Feels like a, a, a,
an agent in that sense is a

human, but just so much faster.

And moreau autonomous in a sense.

Like we've been building dashboards to
look at data for however long because if

you were to look at the raw data, you'd
make the wrong inference or you'd, you

know, I wouldn't be able to write the
sequel in a way that I would be confident

enough to know that the data was right.

So I need a data scientist or, a data
analyst to write, to make sure that

the dashboard is the correct data
and then I can infer a decision or

make a decision about a new product
direction or whatever it is from that

dashboard, because that's trusted.

And it feels like what you've done here
is basically, had to do that because you

have this incredibly fast AI employee that
will just make all the wrong decisions

incredibly quickly and not question them.

So you have to force them into
this box where they're only

looking at this one thing.

so I suppose it's in that sense, it's a
human nature amplified or something, but.

Matt: Yeah, I think one of the
interesting things I suppose is that

I've moved from dashboards to reports.

that kind of overview is really
kind of for me to trust that the

data is flowing in the right way.

Right.

And if like the email is kind of
like the email such as broken, I can

it like blow away the cache, change
the way that comes in, you know,

that's all very deterministic, right?

But for getting a kind of an overview
about what's going on, I don't

rely on anything deterministic.

I get the agent to look at everything
and there's another skill which

is give me my daily report.

And what that does is it can intelligently
prioritise it can take into account kind

of like the severity of the issue, how
long it's been kind of sitting there

is with a customer who's been kind
of messed around before, and so on.

And you know, this, this item's been stuck
in Belgium for like six weeks, right?

It's kind of happened or to go kind
of, oh, there are two parcels which are

meant for this particular destination.

Do they have a kind of a commonality?

are a whole bunch of, deliveries
being caught up in Oakland, right?

And I was able to kind of identify
that, but it's that intelligence that

wouldn't have surface fire, a dashboard,
which kind of like where the structure

presupposes the information, where
what I can do instead because I have

this ground tree, this give the agent
the soup and go, this is the output

report I want about how you feel about
n is gonna be different day to day.

And that's been a really
interesting process as well.

'cause I think the tools
that we use to interact with

agents like this is different.

It reminds me kind of, of how I imagine
offices worked in the 1950s, right?

it's kind of, it's reports based to
somebody at a desk who's responsible

for this information going through.

And I wonder like how much we
have to learn from kind of like

fashioned information, people
flowing offices instead of computers.

Like how we collaborate with these things.

Tom: Yeah, many, so many examples there
that you, of really interesting agentic

use cases that are also beyond code.

It feels like coding, agentic
coding is one true use case that has

really been nailed, but most people.

But I think some of the examples
you gave there are this, almost like

operating system for how to even
build the foundation of a company

or a product or a team on top of, so
lots of really inspiring things there.

I know we are well over time, so we
should probably switch to, final, quick

fire questions if that's okay with you.

Unless there's anything else, any last
thoughts you wanna share on that topic?

AgTech fulfilment overall.

Matt: be remiss if I didn't
put in a couple of plugs.

Is that okay?

Jonny: Okay.

Tom: for it.

Matt: plug number one is there are
gonna be more blocks for sale because

I didn't, I overproduced the batch.

so there are some, which didn't go
to Kickstarter backers, and that is

waiting for me to set up the Shopify.

So please, if you go to Poem Town, there
is a link on that page to Substack.

sign up with a, you'll find out
first about when to get those clocks.

And if you wanna follow along with
inanimate, we have some things coming up.

go to news tech and subscribe,
like, and subscribe.

As I say, like we're beginning to be a bit
more vocal about what we're doing right.

And, that's, an interesting story as well.

So I hope people enjoy
following along with that.

Jonny: Fantastic.

I'll have to have you back in
a year or, something and, see

what's happened with Inanimate.

It'd be awesome.

Matt: Would love to.

Now, I've been looking forward
to these quickfire questions.

Jonny: Oh God.

Matt: on, 10 hooks about this.

Tom: Okay, let's kick off.

Maybe we could do a couple each,
Jonny, and then wrap up there.

So I, we'll kick off with, when
a GI takes over everything.

What will you do with all
of your free time Matt?

Matt: You know what?

Probably exactly what I'm being I'm doing
now, it will just feel like a very kind

of like pursuit, bit like kind of like
hiding in the woods and doing the things

that the machines could be doing anyway.

Jonny: Them,

Matt: put, like, why do
people still play chess,

Tom: Of that.

Matt: question, right?

Like,

Jonny: yeah.

Matt: are way better at
chess and like, what?

Why did it, because it's fun.

So I think I get very fulfilled out
of what I'm doing at the moment.

So yeah, probably the same.

Jonny: Feel like that gives you,
hope for the human condition or that

we will find fulfilment generally
as a species because it's not just

all about efficiency at all costs.

Matt: Yeah, I'd like to
be optimistic about that.

I hope so.

Jonny: Yeah.

Yeah.

so my, quick fire question is, if
there's one product that you wished

you'd created physical, digital
in the world, what would that be?

Matt: I am, this is not the
product, it's the creative.

there's the product friend friend.com,

which is the AI pendant
where you speak to it.

And now with version two,
it kind of speaks back.

And, I think of the, there's a few
kind of overlooked areas and ai

and one of them is kind of how we
use AI as a kind of a, a companion.

Like it's only been done in the kind of
a, a game, like a playful way so far,

but nobody's taken it really seriously.

Friend is doing that and their
recent creative, which was all you

know, analogue v analogue sound,
and tiny vignettes, which showcased

the product but don't centre it.

They're more of the lives of
people around it is just inspiring.

It's absolutely beautiful
and absolutely pitch perfect.

Jonny: It must because they spent
a lot of money on friend.com.

That's, my main memory of friend.

I don't have a friend.

I mean, if I have some
friends, I don't have a friend.

Always a very difficult thing to say.

I assume when you're talking
about, Tom, over to you.

Tom: And my last question, you've
given us a tonne of examples today of

things that you have built recently,
but I have to ask you what the silliest

thing is that you've prototyped like
with ai, purely because you could.

Matt: There have been so many ridiculous
projects, but they often, they predate ai.

I think my, one of the hardest projects
was I made a vending machine to sell books

that would tweet when it sold the book.

was great.

I took it on tour around like various,
kind of like lobbies in London that

was in like it was in Google for that.

And, it would they would stock
books that were recommended by

people working in the building.

It weighed 270 kilogrammes.

It was almost impossible to move.

My house was full of vending
machine parts for years, and I

was like, do not pick up hobbies,
which like won't sit through doors.

It's a, a ridiculous tube.

But that was pre ai, so I'm not allowed,
I'm not allowed to have that one.

But yeah, like the in the string of
very many ridiculous perceived selling.

Jonny: Amazing.

I think you're allowed that.

I don't think, I don't think
there's an AI constraint on this.

I think you're allowed
your vending machine, Matt.

Tom: Yeah, definitely.

Although I am curious to ask the
follow up question of if you're

doing that today, is there any
aspect of that AI helped you with?

AI unlocked the possibility
of doing another

Jonny: Forward, move the vending machine.

Tom: around London?

Matt: Yeah, it would've
made technology much easier.

So vending machines internally have
a protocol, which is a bit like USB

called, MDB, I think was invented by
Coca-Cola, Japan in the mid 1970s.

And the technology spec is like 300 pages.

It was very difficult to integrate with.

I think that it would make much easier.

but you know, it wouldn't
help with the other problem.

The vending machine, which was his thing
to tell you about vending machines, right,

is they're fridges basically, right?

So if you turn the refrigeration
element off, 'cause it's noisy, what

happens is it just gets and warm.

The glue to hold my shelf talkers
on, which is the technical name for

the cards underneath the books that
say like why they're good started

melting and all the shelf talkers
would fall off after about three days.

Finding a formulation of
adhesive to stop them falling off

took me weeks,

side the seller tape.

If you're interested the best stuff,

Jonny: Go.

Matt: to a dozen others
before I landed on that.

AI couldn't have helped with that.

So there's, you know, there's utility
for us humans yet, that's what we can do.

Jonny: The meat layer.

Yeah.

Tom: Love that.

Love

Jonny: right.

We should, finish there.

Matt, thank you so much for joining us.

This is, a monster edition and so glad it
was learned so much and really enjoyed it.

So, maybe we'll have you back.

I would love for that to be true.

but in the meantime, best of luck
with inanimate and and thank you.

Matt: Jonny Tom, thanks for having me.

Been a pleasure.