_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

The first web browser was read/write, a fact brought to my attention by my colleague Stephen Hood: > Berners-Lee’s original Web browser running on NeXT computers showed his vision and had many of the features of current Web browsers. In addition, it included the ability to modify pages from directly inside the browser – the first Web editing capability. (home.cern) The original promise of the web was a fully-distributed amalgamation of individuals with web servers. Unfortunately, running your own server never got mainstream traction, and my guess is that this is largely because it never became an easy thing to do. So, platforms took over, streamlining the process of publishing to the web, and thereafter enshittifying it. What's the solution? An easy way to truly host your own content (like if every router, a constantly-connected device, had its own reliable web front-door)? A distributed platform where everyone becomes a host, through peer-to-peer networks overlaid on the internet? Activity Pub somehow? There has to be a better way.

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

The first web browser was read/write, a fact brought to my attention by my colleague Stephen Hood: > Berners-Lee’s original Web browser running on NeXT computers showed his vision and had many of the features of current Web browsers. In addition, it included the ability to modify pages from directly inside the browser – the first Web editing capability. (home.cern) The original promise of the web was a fully-distributed amalgamation of individuals with web servers. Unfortunately, running your own server never got mainstream traction, and my guess is that this is largely because it never became an easy thing to do. So, platforms took over, streamlining the process of publishing to the web, and thereafter enshittifying it. What's the solution? An easy way to truly host your own content (like if every router, a constantly-connected device, had its own reliable web front-door)? A distributed platform where everyone becomes a host, through peer-to-peer networks overlaid on the internet? Activity Pub somehow? There has to be a better way.

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

The first web browser was read/write, a fact brought to my attention by my colleague Stephen Hood: > Berners-Lee’s original Web browser running on NeXT computers showed his vision and had many of the features of current Web browsers. In addition, it included the ability to modify pages from directly inside the browser – the first Web editing capability. (home.cern) The original promise of the web was a fully-distributed amalgamation of individuals with web servers. Unfortunately, running your own server never got mainstream traction, and my guess is that this is largely because it never became an easy thing to do. So, platforms took over, streamlining the process of publishing to the web, and thereafter enshittifying it. What's the solution? An easy way to truly host your own content (like if every router, a constantly-connected device, had its own reliable web front-door)? A distributed platform where everyone becomes a host, through peer-to-peer networks overlaid on the internet? Activity Pub somehow? There has to be a better way.

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

The first web browser was read/write, a fact brought to my attention by my colleague Stephen Hood: > Berners-Lee’s original Web browser running on NeXT computers showed his vision and had many of the features of current Web browsers. In addition, it included the ability to modify pages from directly inside the browser – the first Web editing capability. (home.cern) The original promise of the web was a fully-distributed amalgamation of individuals with web servers. Unfortunately, running your own server never got mainstream traction, and my guess is that this is largely because it never became an easy thing to do. So, platforms took over, streamlining the process of publishing to the web, and thereafter enshittifying it. What's the solution? An easy way to truly host your own content (like if every router, a constantly-connected device, had its own reliable web front-door)? A distributed platform where everyone becomes a host, through peer-to-peer networks overlaid on the internet? Activity Pub somehow? There has to be a better way.

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

The first web browser was read/write, a fact brought to my attention by my colleague Stephen Hood: > Berners-Lee’s original Web browser running on NeXT computers showed his vision and had many of the features of current Web browsers. In addition, it included the ability to modify pages from directly inside the browser – the first Web editing capability. (home.cern) The original promise of the web was a fully-distributed amalgamation of individuals with web servers. Unfortunately, running your own server never got mainstream traction, and my guess is that this is largely because it never became an easy thing to do. So, platforms took over, streamlining the process of publishing to the web, and thereafter enshittifying it. What's the solution? An easy way to truly host your own content (like if every router, a constantly-connected device, had its own reliable web front-door)? A distributed platform where everyone becomes a host, through peer-to-peer networks overlaid on the internet? Activity Pub somehow? There has to be a better way.

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper

Let's start with a fictional case study. This one's inspired by user research I've been doing into human interactions with language models. > Cynthia is planning for her first big solo trip after college: a visit to Australia. It's been on her bucket list since she was a kid, but she knows very little about where to go, what to do, or how to get around. She starts a conversation with a chat fine-tuned large language model, and discusses which cities to visit based on her interests and budget. > > The model suggests safe parts of town to stay in, and they discuss whether to choose a motel or AirBnb based on her budget. As the trip approaches and her conversations continue, the model writes her a to-do list of things to buy – chargers, spare battery packs, a basic medical kit, a pillow for the plane, headphones, and adaptors – and gives her tips for long plane journeys. In the end Cynthia has a hugely successful trip, and visits areas and towns that she never would have thought to seek out thanks to the advice of the model. This is a new sort of interaction with computers, one that's far more personal and flexible than anything in the history of human-computer interaction. It's something I've been calling "thought partnership". > **Thought partnership.** A relational interaction with a computer system in which the user and computer collaborate to uncover information & insights relevant to the topic of conversation, the user's interests, and their values. This term has become a handy concept handle for me in the months since I started using it, so I thought it was about time I shared it. Let's unpack it a bit more. Thought partnership has a couple of unusual traits: * **It's open-ended.** Thought partnership is an exploration, you start it and you're not sure where it's going to take you; at best it's novel & surprising. * **It's compounding, not transactional.** Unlike search, thought partnership involves an ongoing back-and-forth between the computer & user (in text or other media), to develop ideas and thoughts. * **It's personal.** Good thought partnership requires a contextual understanding of the user, their current state, feelings, values, and requirements, in order to tailor the outputs to them. Recently, I've been experimenting with making a tool-for-thought-partnership. For the moment it's quite simple: a language model that has access to a read/write text buffer. The buffer acts as a kind of long-term memory scratchpad.[1] Using this buffer, the model can preserve details about me over the course of many interactions. In addition, I can inspect what the tool understands about me, and even write something directly into it to inform the model about me, or how I want it to behave. Already, the simple combination is having some novel effects. Since I've been using this tool it's: * Recommended me a restaurant while travelling in San Francisco based on its understanding of my taste & values. It also suggested a specific meal to try, which was on the menu (it was delicious). * Encouraged me to go to sleep at a particular time on my return to Sydney, as it recalled that I would be jet-lagged. * Checked in about this blog post to ask how the writing process was going, given I'd mentioned it a few days prior. Now, of course the engine powering all this is a language model. But better tools for thought partnership aren't just about better models. The best way to improve this interaction is to build more thoughtful connective tissue – that is, layers of software that can read from, write to, and search stores of information to gain a deeper context, as well as orchestration layers that can trigger API calls and tool use. We're still starved for good innovation on the context layer, in my view. If you want to play around with the tool I made, you can access Familiar on mobile, Mac or web here. * * * 1. Another common method for retaining memory is vector search over message history. The buffer has some advantages: it contains summarised insights that can be prepended to every prompt, not just those that relate to the latest message. These insights can also evolve over time as the model revised them. Pairing the two methods together would be even more powerful. ↩︎

> If you haven’t used an app in a long time it should start collecting dust as a suggestion to delete it pic.twitter.com/7VyUBmp0a3 > > — Neil Sardesai (@neilsardesai) April 3, 2021 We often think things like "dust" or "decay" as downsides of the physical world. One big benefit of cloud computing is that nothing decays – your data is stored forever, ethereally, somewhere _out there_. But decay, dust, disintegration — they are all vital to understanding the state of our physical world. * * * Back in 2019, when I was working at Google, I was taking a walk down one of the wharves alongside our Sydney office. I came across a portion of the boardwalk that had just been replaced. Planks of wood have this great latent data-visualisation feature: by default, they display how old they are, and indicate when they need replacing or updating. I thought – imagine if web content did that? To a certain extent, of course, it does already. Links "rot" (fantastic terminology btw), designs go off-trend and then weirdly become hip again. But I'm interested in how intentional decay & dustiness has made its way in our interfaces, how incorporating a very anti-tech part of the natural world can make our computer interactions more sensible to humans. * * * A few examples I've stumbled across over the years: **1)** Web developer Chris Coyler wrote a wonderful article about how to show readers that a piece of documentation was ageing and no longer relevant. One of his solutions: having pages without recent updates start to decay organically over time, with words blurring and eventually falling off. **2)** Artist Zach Gage's Temporary.cc (now fully degraded – it's just a blank page), is another take on this idea. But instead of decaying over time, his website decays with use, like an old shoe that starts to fall apart on a long hike. From Zach's portfolio: > Temporary.cc eschews this paradigm. For each unique visitor it receives, Temporary.cc deletes part of itself; a single character from its own code. These deletions change the way browsers understand the website's code and create a unique (de)generative piece after each new user. Because each unique visit produces a new composition through self-destruction, Temporary.cc can never be truly indexed, as any subsequent act of viewing could irreparably modify it. **3)** The idea of collecting dust has been used in mainstream tools like Trello for a while, with cards crumbling and ageing when they haven't been used in a while. **4)** Nattyware had a prototype app back in 2004 that covers all unused items in the desktop with a fine layer of dust. **5)** A new read-it-later app Alfread asks you to blow off the dust from articles you've had on your digital to-read pile for too long: > From dust to Alfreadust. > > Now, when you open an article that was saved a long time ago, it will be covered in dust. > > If you still want to read it, blow into the mic to clear it off: pic.twitter.com/7t3zXEkT7t > > — Alfread (@AlfreadHQ) April 1, 2021 * * * Our physical world is the original interface. It makes sense that even the less-technologically-savoury parts of it give us important cues about what's going on. Let's bring the messy, mouldy, decaying bits of real life into our digital world. Dust is dead, long live dust.

_Or, "Down with the hierarchy-archy"_. Let's talk about the filesystem. Somewhere deep in MIT, around 1965, a group of researchers were putting the finishing touches on their new computer operating system, _Multics_. One of the breakthrough features? A "hierarchical, location-addressable" filesystem. Instead of dealing with an unstructured list of data on your computer, you could collect data in "files" and put them in "folders" — in other words, your information was structured in a _hierarchy_. And, to interact with some data, you'd just need to know where it was in the hierarchy – a _location address_. Fast-forward four years to 1969, and a team of breakaway Multics engineers invented Unix, carrying the files-and-folders metaphor with them. Today, with the huge influence Unix had on the genesis of personal computing, the hierarchical location-addressable filesystem is everywhere. And it's _fine_...? Well, depends on who you ask. In my view, the files-and-folders metaphor has got us all in a conceptual rut. Real files and folders—the ones our filesystem metaphor are based on—they're physical objects. And physical objects have certain constraints, which we have unwittingly enforced upon every piece of data in every computer in the modern world. For example, the research document I've put together writing this article applies to a bunch of other work I have on at the moment. But it can only _live_ in this folder with my newsletter articles – one folder, one file, one physical place _(copy... paste...)_. I also have a bunch of information on my computer that I'd love to connect. A PDF article I downloaded the other day on David Lynch's maps of Twin Peaks would pair excellently with a book I saw the other day on cartography, and with a _specific paragraph_ in a journal entry on travel I made in 2015, and with an email from a friend who's obsessed with pop culture. But again, I confont this clunky metaphor – if I want to link files, I need to refer to where they are, and I can only refer to the files in their entirety. If I move the files, change computer, or email a file to a friend, all these relationships are gone. Poof. That's not to mention needing to open specific apps to see what's inside my files, the lack of easy tagging and metadata, the hell that is broken links on the web... Because of the limitations of filesystems for actually organising information in a way people find useful, proprietary layers of software have been built to abstract away from all this – think Notion, Airtable, Google Drive, Evernote. Even tools like iTunes are just databases superimposed on a folder full of files to make them palatable. More recently, apps like Hook have emerged that explicitly link files with unique IDs instead of location addresses. While it's a fascinating take, it's still an extra layer of brittle complexity – and you'd better hope the company never goes bust. But there have been glimpses of a better future. For example, WinFS was an ambitious (but doomed) filesystem pioneered by Microsoft in the early 2000s. It treated your entire filesystem like a relational database – instead of having discrete closed-off files, information was set free like a spreadsheet. You could link anything to anything else, sort and filter by rich metadata, and form relationships between different bits of content. _(When it fell over, Bill Gates referred to it as his "greatest mistake", saying the world was not ready for it yet and promising its return)._ At the moment, I'm most excited by content-addressable filesystems, which also originated in the 60s but then somehow slipped away unnoticed. In a content-addressable filesystem, instead of pointing to _where_ a file is on a computer, you point to _what_ it is. Think of it like someone asking you what you want to eat for dinner. Instead of saying "fridge, third shelf, 20cm to the left" (a location address) we'd probably rather say "the salmon fillets, please" (a content address). The latest take on mainstream content-addressed storage is the Inter-Planetary File System (IPFS). It's not just concerned with having an address for the files on _your_ computer, but also files across the _entire internet_. Each file uniquely identified in a "global namespace". Importantly, IPFS needs no centralised host, no web server, no external company like Dropbox or iCloud Drive – data is distributed over all the computers in the network. And if you link to something, that link won't break with a reshuffle of your files, or a website going offline – so long as the data can be found somewhere in the world (by matching its _content_ , not its location) then everything will work. _(It reminds me a bit of the original vision forProject Xanadu, another piece of brilliant information technology archaeology, a predecessor of the web... well, in their words, the web "trivialises" their project, with "ever-breaking links" – I recommend you look it up)._ I'm convinced that the fact we stick all our digital information in "files and folders"—metaphors for actual physical locations—restricts our imagination about what we can actually do with our data. I'm also convinced that there's a plethora of once-helpful, decades-old digital metaphors, which are now ripe to be picked, composted and replaced with something new. What these metaphors are, is left as an exercise to the reader. ⌘ **Some noteworthy links from my research:** * A fascinating paper titled "Hierarchical Filesystems Are Dead" (for a nuanced take...) * Wired Magazine on Project Xanadu * "Hypermedia" * Lotus Improv (Wikipedia) > "At the same time, the new product would allow users to group data "by purpose", giving it a name instead of referring to it by its position in the sheet. This meant that moving the data on the sheet would have no effect on calculation. * Ideal OS, a vision for a rethinking of the OS * Spec for a proto-digital filesystem made with paper