It’s Sunday, 7:30 pm. You want to enjoy the last few minutes of the weekend but instead you’re typing the letters t o i l e t p a p e r into a search bar. You watch TV for a bit and then look down to see a grid of different kinds of toilet paper with pictures. You scroll. Some are 1ply, others 2ply. There’s a 2 for 1 deal on a 9pk, but is that cheaper than the 18pk from the other brand? You briefly try working it out before hitting the add to cart button with reckless abandon. A spinner shows. It goes away and you see another button “quantity: 1 - add to cart.” You click this button. A spinner shows again. You watch TV for a bit. You look down to see a green checkmark. You tick off toilet paper and start typing the letters “m i l k”...
It only takes 20 minutes to finish your list and you’re grateful you have the luxury of being able to spend the 20 minutes getting whatever you want. Yet, you’d absolutely get someone to do it for you if you could.
I’m going to call the time we spend on tasks like this, stupid minutes. That is time spent on tasks which: (1) aren’t ends in and of themselves but merely means to ends. (2) a machine could cheaply do them as well as you. (3) And yet you're the one doing it. The stupidness of stupid minutes is not inherent in the task. Rather it’s that the gap between technology we’ve created and your access to it, is stupid. So buying toilet paper in 2022 wouldn’t have been stupid minutes, because we didn’t have a cheap machine that could do it as well as you, but it is in 2026.
There are stupid minutes everywhere you look. I’m releasing a thing to fix some of them, and I’m starting with the stupid minutes spent on shopping. Specifically shopping in South Africa. Specifically Shopping in South Africa, at Woolworths for many things. It’s called Pelicart and you can now join the beta. You message Pelicart over whatsapp and it securely uses your woolies dash account to do one of three things. Search, add or remove from your cart. It does this while you do other things. When you message Pelicart it’s like messaging someone at a store who you’ve hired or begged to do your shopping. You can send pelicart a handwritten shopping list an email or a recipe.
About two minutes later everything you asked for will be in your cart and this is where Pelicart stops. You can check it got the right stuff, make some adjustments if needed and checkout of the real woolies dash app like you always do.
Disposable Programs and the End of the App
I see artists vowing to never use the technology that makes Pelicart possible as an ethical principle, in the same way vegetarians vow to never eat meat. I see programmers who embrace it unconditionally in the same way some people only eat meat. Unfortunately, I don't have all the answers to what we should and shouldn't use this technology for. But I don't have zero answers either. I have exactly one answer which I'm quite sure is correct: AI should be used to buy us toilet paper.
For some people this has never been a problem. At a certain level of wealth you stop having to think about buying toilet paper. You have a PA take on the responsibility, decide which toilet paper to get and buy it for you and so you spend zero time thinking about or buying toilet paper (or you get a bidet from Japan but just pretend those don't exist).
Up until recently you’d have needed a lot of wealth to be one of these people. This stopped being the case about 24 months ago, at which point many more people could have become one, if we wanted them to. It doesn't end at toilet paper. Like papercuts, stupid minutes bleed our time. Filling in pdf forms by hand. Booking meeting rooms in your office. Typing your ID number to open an email. These are stupid things we could have been getting computers to do for us but haven't. And I think that’s bad. You might take a zen approach to this and regard these stupid minutes as being not necessarily stupid but rather an experience of life to be present for that’s no less valid than watching a sunset or driving a car or anything else. My answer to that is mu.
Not only did we make humans keep spending stupid minutes, in some cases we used computers to purposefully create even more stupid minutes. The time it takes to find your phone so you can click approve on a $2 purchase, are each one of them stupid minutes. The total amount of time wasted on getting humans to approve obviously legitimate transactions is disgusting to me. And what's even more disgusting is that we've somehow convinced people that approving transactions is a good thing for humans to be doing with their time, as though any increase in bank safety is justified even if it costs us collectively hours of our lives for like a 0.1% reduction in the probability of fraud occurring. I’m not saying that is the actual number but we don’t know what the number is and even if we did couldn’t turn off two factor auth, and accept the risk. From the bank’s perspective you not only will get the maximum amount of security but you ought to want it too. Which makes sense, why would the bank consider your time a cost.
I don't know why we're here, I don't know why you're reading this, I don’t know what you have to do to achieve living your life well, but I suspect it’s not comparing the price of toilet paper.
"Everyone is hurting each other, the planet is rampant with injustices, whole societies plunder groups of their own people, mothers imprison sons, children perish while brothers war."
The Goddess answers: "What is the matter with that, if it's what you want to do?"
AI should be making our lives easier. In many ways it has, but we should be seeing the total stupid minutes spent by people on the planet dropping to zero. In my estimation the AI we had two years ago was sufficiently powerful to do this. But when I look at my family and friends, I see them spending, if anything, more stupid minutes. Sixty60 just added an AI assistant called pixie which is so stupid I can’t bring myself to capitalize it. Does pixie stop you from having to compare the prices of toilet paper as it so easily could? No it’s tinder for bread at the bottom of your screen.
It might seem like i’m frustrated at the fact that people spend any time on chores like shopping and emails, i’m not. I’m frustrated that there is so much low hanging fruit to make people’s lives significantly better, but no one is picking it. Discovery bank has been categorising my purchases very nicely, but I still have to spend 3 minutes entering several different numbers to send money to someone. I'm not saying this is like a huge issue or that I'm mad about losing minutes of my life when I watch The Vampire Diaries for several hours. But it’s a huge issue that we have the ability to remove annoying tasks from so many people’s lives and haven’t and i’m mad that we don’t seem to be.
The reason for this is that we're in the horseless carriage phase of AI. Before cars were invented, you saw things like this:
I am far from the first person to say that this is what some AI tools are like these days. My favorite essay about this is https://koomen.dev/essays/horseless-carriages/ in which Pete Koomen shows how Gemini has been integrated with Gmail in exactly the same way the engine has been integrated with wheels, in the above picture.
The point is that in a horseless carriage period we are limited by our beliefs of what problems exist, what technology can solve and in what ways it can solve it. When you see an engine you see something to make your carriage be horseless, instead of a car. When we see AI we think of making something to make our apps better instead of... well we don’t know yet.
Thinking about shopping and banking and the like in terms of apps and a fixed series of actions people want to use them for, is the problem. The actions you can take in an app are means not ends. But we’ve been using them for so long we have started thinking about them as ends. Categorising the transactions I make so I can look at them neatly is a waste of time if i effectively have an accountant that can interpret them without me ever looking at them. I don’t look at lists of transactions to scroll through them, I look at lists of transactions as a means to answering questions like what have I been spending money on, how much money do I have, and of course; oh boy did I really spend that much?? We should not be thinking about how to make existing apps like Notion or Monday or Asana better with AI. Rather we should be thinking about if we even still have the problems they were designed to fix.
This brings me to why I'm writing this article today. The Sixty60 designers didn't build their app as a giant text box where you'd have to type out commands to buy milk on your phone (with curl). That wasn't because it was impossible, it was because (1) most people can't write code, and (2) even if you could, typing out a command for each item would take far longer than just tapping a button. So they did what every shopping app does: they built a screen with a search bar and buttons for adding and removing things from your cart.
What if though, you not only knew how to program, but also knew how to program as well as an experienced developer and could do it faster than any developer on Earth, all without having to pay the large salary such a person would command, if they could even exist. Well if this were true, then you would be in our current reality. And in our current reality the assumptions that drove Steve Jobs toward touch screens and fingers no longer hold.
Until now, the vast majority of computers have allowed us to do things we find valuable, by showing a predefined set of actions with which we have the power to compose in a specific order to attain a valuable thing. In a way each button is like a piano key, but you still have to play them correctly to make music. For example, here are the predefined set of actions i can see in google docs right now
Here are the predefined set of actions you can take in the woolworths app when searching for toilet paper:
And here are the predefined set of actions in SPSS
We’ve gotten so used to expressing our desires by composing button clicks that it’s easy to conflate the button itself as thing we desire. When really, they are just how we have converged on representing the actions you can take. They are not the only and as of recently not even the best way to achieve our actual ends which are buying milk or a making a graph.
Interacting with computers mediated by buttons makes sense only if the person looking at the screen is the only thing that can decide how to compose actions to accomplish exactly what they want. For the longest time this was true, today though, this is no longer true because for most things, sonnet 4.6 can at the same time select the correct actions and express those actions as code, faster than you can express them with a mouse or your finger. The only problem we have then is how to expose actions that were previously buttons, to AIs. Weirdly we already have. What we need exists (metaphorically) under the button. It’s the code human developers have spent all their time writing for the last 50 years. As of today, most of this code is only designed to be reached by humans clicking buttons. The actual code that runs when you click the button, exists though, and If an AI had access to it, it could take actions on behalf of a person much faster. Mostly, AIs don’t yet have the ability to interoperate with the code under buttons, so they can neither see the actions nor execute the actions an app can perform, even though it is intelligent enough to both understand what we want and reify it using those actions.
If the problem isn’t AI intelligence then do we just have to find a way for AIs to be able to communicate with the programs we care about on our behalf? Basically yes, we once thought the answer to this was something called MCP, but this doesn't seem to be the case anymore so we’re not gonna talk about it. Instead it's command line interfaces which are proving to be the best way for AIs to do things on your behalf. In an ironic twist of fate, the mouses and windows Steve Jobs borrowed from Xerox to replace the command line are now themselves being replaced by command line. Command line applications are just programs that you interact with via text. For example here’s a command line application called yt-dlp that lets you download youtube videos:
If you type:
It will download the video for you. Easy. Turns out, modern AI’s are really good at writing commands like this to achieve things on your behalf because it’s just text.
For the last few months it's been clear to us that AIs are exceptionally good at programming. You’re probably imagining this means an AI making stuff like websites or apps that people would then use. Wall street certainly thought this a few months ago, to the extent that many companies which only make money by selling a single website or app, suddenly became less valuable. It is true that AIs are exceptionally good at building traditional software like this, but they’re equally good at a category of programming we don’t have a name for because it doesn’t fit with our fundamental assumptions about what programming is and should be for. This kind of programming is one where the program itself has no value, only the results it generates do. They are programs which are entirely customized to your very specific task and deleted instantly after they’ve been run. For example, you ask claude if your psychiatrist emailed you, claude writes a full Python program to search gmail. It then executes it, gets the result and notices it only got back 30 emails. So it writes another full Python program, this time adjusting the number from 30 to 1000 and based on that result replies “yes.” From your perspective this took 10 seconds and all you saw were the words “thinking…”
These are disposable programs. The point of these programs is just the result. It’s kind of like when you’re using a calculator to do your taxes, you input some numbers, get a result which you write down and move on. Once the AI has the result, it doesn't matter what happens to the program and something that would have taken a human days to write will be executed then deleted with the same care as an accountant pressing AC on a calculator. We were born with programs being something complicated to create, something that is impressive when done well. When we say AIs go beyond the power of human programmers, not only are they superior at the normal type of programming we’ve grown up with. They are also superior at programming in ways we didn’t know you could program.
Today, there are probably some problems only a team of cracked developers can solve. Perhaps Opus 4.6 is worse at coding than such a team. But for everything else Opus 4.6 will do the exact same quality of work, in seconds over and over again for the entire night. This change in quantity is also a change in quality. Developers paying hundreds of dollars for a Claude Code subscription which they then use to make a product for a SaaS company or add features to one is what strapping an engine to a carriage looks like. If you have access to something that knows how to program as well as an experienced developer, why do we need the saas app or feature anyway?
Wall street worried that companies would vibe code their own version of trello and cancel their subscriptions, what would be worse is companies not needing to vibe code trello at all, because agents can manage their own tasks better with python and tell you what they’re working on in English.
When GPT 5 came out, I got it to do my mom's shopping. She could send a picture of a handwritten list to a whatsapp number, GPT would search then add groceries to her cart using disposable scripts but I stopped working on it, it had some bugs and I got demotivated when ycombinator rejected it.
Whenever I’ve seen my mom shopping since then, I’ve felt deep guilt because I didn’t make this fully usable. I felt even more guilty when sonnet and opus 4.6 came out because I knew that they would absolutely nail this task even better than GPT 5. Part of me hoped or assumed that someone would do this for me when openclaw got big. But no one ever did. Part of me hoped or assumed the companies themselves would do it, but checkers made bread tinder. It's become clear to me that stupid minutes won't go away on their own. But they will go away the when we decide they should. A tiny amount of time spent opening doors will yield huge returns in our every day lives, because the same intelligence currently building apps in one shot can more than easily do our sludgework, we just have to let it.
I don't know how much we should use AIs for creating art or writing or how we should aesthetically value what they do. I do know it's pointless to argue about this when there's a million things we obviously should be using AIs for that we aren't. And I do know that it's better for humans to spend more time painting and less time comparing toilet paper sales. So that's where I'm starting.
Pelicart Details
Pelicart will be free in beta. Once it seems to be working well with just Woolies Dash I'll invite more people to the beta, probably start charging something and eventually add Sixty60 and PnP and Dis-Chem, so you can genuinely and completely never have to spend stupid minutes on shopping again. Pelicart isn't designed to replace browsing through stores. A couple of days ago I was hungry and tried using Pelicart to buy me some snacks. It sucked at that. I found it way better to just scroll through the app.
Pelicart is also just step 1 for me. I mentioned some other types of stupid minutes earlier like two factor auth, which I think are just as stupid and which I think are just as easy to do away with and which I will do what I can to help do away with.
While writing this I got an email from the read it later app Matter which looked like this and I think it perfectly sums up the direction I see computing going:
It’s Sunday, 7:30 pm. You want to enjoy the last few minutes of the weekend but instead you’re typing the letters t o i l e t p a p e r into a search bar. You watch TV for a bit and then look down to see a grid of different kinds of toilet paper with pictures. You scroll. Some are 1ply, others 2ply. There’s a 2 for 1 deal on a 9pk, but is that cheaper than the 18pk from the other brand? You briefly try working it out before hitting the add to cart button with reckless abandon. A spinner shows. It goes away and you see another button “quantity: 1 - add to cart.” You click this button. A spinner shows again. You watch TV for a bit. You look down to see a green checkmark. You tick off toilet paper and start typing the letters “m i l k”...
It only takes 20 minutes to finish your list and you’re grateful you have the luxury of being able to spend the 20 minutes getting whatever you want. Yet, you’d absolutely get someone to do it for you if you could.
I’m going to call the time we spend on tasks like this, stupid minutes. That is time spent on tasks which: (1) aren’t ends in and of themselves but merely means to ends. (2) a machine could cheaply do them as well as you. (3) And yet you're the one doing it. The stupidness of stupid minutes is not inherent in the task. Rather it’s that the gap between technology we’ve created and your access to it, is stupid. So buying toilet paper in 2022 wouldn’t have been stupid minutes, because we didn’t have a cheap machine that could do it as well as you, but it is in 2026.
There are stupid minutes everywhere you look. I’m releasing a thing to fix some of them, and I’m starting with the stupid minutes spent on shopping. Specifically shopping in South Africa. Specifically Shopping in South Africa, at Woolworths for many things. It’s called Pelicart and you can now join the beta. You message Pelicart over whatsapp and it securely uses your woolies dash account to do one of three things. Search, add or remove from your cart. It does this while you do other things. When you message Pelicart it’s like messaging someone at a store who you’ve hired or begged to do your shopping. You can send pelicart a handwritten shopping list an email or a recipe.
About two minutes later everything you asked for will be in your cart and this is where Pelicart stops. You can check it got the right stuff, make some adjustments if needed and checkout of the real woolies dash app like you always do.
Disposable Programs and the End of the App
I see artists vowing to never use the technology that makes Pelicart possible as an ethical principle, in the same way vegetarians vow to never eat meat. I see programmers who embrace it unconditionally in the same way some people only eat meat. Unfortunately, I don't have all the answers to what we should and shouldn't use this technology for. But I don't have zero answers either. I have exactly one answer which I'm quite sure is correct: AI should be used to buy us toilet paper.
For some people this has never been a problem. At a certain level of wealth you stop having to think about buying toilet paper. You have a PA take on the responsibility, decide which toilet paper to get and buy it for you and so you spend zero time thinking about or buying toilet paper (or you get a bidet from Japan but just pretend those don't exist).
Up until recently you’d have needed a lot of wealth to be one of these people. This stopped being the case about 24 months ago, at which point many more people could have become one, if we wanted them to. It doesn't end at toilet paper. Like papercuts, stupid minutes bleed our time. Filling in pdf forms by hand. Booking meeting rooms in your office. Typing your ID number to open an email. These are stupid things we could have been getting computers to do for us but haven't. And I think that’s bad. You might take a zen approach to this and regard these stupid minutes as being not necessarily stupid but rather an experience of life to be present for that’s no less valid than watching a sunset or driving a car or anything else. My answer to that is mu.
Not only did we make humans keep spending stupid minutes, in some cases we used computers to purposefully create even more stupid minutes. The time it takes to find your phone so you can click approve on a $2 purchase, are each one of them stupid minutes. The total amount of time wasted on getting humans to approve obviously legitimate transactions is disgusting to me. And what's even more disgusting is that we've somehow convinced people that approving transactions is a good thing for humans to be doing with their time, as though any increase in bank safety is justified even if it costs us collectively hours of our lives for like a 0.1% reduction in the probability of fraud occurring. I’m not saying that is the actual number but we don’t know what the number is and even if we did couldn’t turn off two factor auth, and accept the risk. From the bank’s perspective you not only will get the maximum amount of security but you ought to want it too. Which makes sense, why would the bank consider your time a cost.
I don't know why we're here, I don't know why you're reading this, I don’t know what you have to do to achieve living your life well, but I suspect it’s not comparing the price of toilet paper.
There's an amazing quote from the essay Meditations on Moloch:
"Everyone is hurting each other, the planet is rampant with injustices, whole societies plunder groups of their own people, mothers imprison sons, children perish while brothers war."
The Goddess answers: "What is the matter with that, if it's what you want to do?"
Malaclypse: "But nobody wants it! Everybody hates it!"
Goddess: "Oh. Well, then stop."
AI should be making our lives easier. In many ways it has, but we should be seeing the total stupid minutes spent by people on the planet dropping to zero. In my estimation the AI we had two years ago was sufficiently powerful to do this. But when I look at my family and friends, I see them spending, if anything, more stupid minutes. Sixty60 just added an AI assistant called pixie which is so stupid I can’t bring myself to capitalize it. Does pixie stop you from having to compare the prices of toilet paper as it so easily could? No it’s tinder for bread at the bottom of your screen.
It might seem like i’m frustrated at the fact that people spend any time on chores like shopping and emails, i’m not. I’m frustrated that there is so much low hanging fruit to make people’s lives significantly better, but no one is picking it. Discovery bank has been categorising my purchases very nicely, but I still have to spend 3 minutes entering several different numbers to send money to someone. I'm not saying this is like a huge issue or that I'm mad about losing minutes of my life when I watch The Vampire Diaries for several hours. But it’s a huge issue that we have the ability to remove annoying tasks from so many people’s lives and haven’t and i’m mad that we don’t seem to be.
The reason for this is that we're in the horseless carriage phase of AI. Before cars were invented, you saw things like this:
I am far from the first person to say that this is what some AI tools are like these days. My favorite essay about this is https://koomen.dev/essays/horseless-carriages/ in which Pete Koomen shows how Gemini has been integrated with Gmail in exactly the same way the engine has been integrated with wheels, in the above picture.
The point is that in a horseless carriage period we are limited by our beliefs of what problems exist, what technology can solve and in what ways it can solve it. When you see an engine you see something to make your carriage be horseless, instead of a car. When we see AI we think of making something to make our apps better instead of... well we don’t know yet.
Thinking about shopping and banking and the like in terms of apps and a fixed series of actions people want to use them for, is the problem. The actions you can take in an app are means not ends. But we’ve been using them for so long we have started thinking about them as ends. Categorising the transactions I make so I can look at them neatly is a waste of time if i effectively have an accountant that can interpret them without me ever looking at them. I don’t look at lists of transactions to scroll through them, I look at lists of transactions as a means to answering questions like what have I been spending money on, how much money do I have, and of course; oh boy did I really spend that much?? We should not be thinking about how to make existing apps like Notion or Monday or Asana better with AI. Rather we should be thinking about if we even still have the problems they were designed to fix.
This brings me to why I'm writing this article today. The Sixty60 designers didn't build their app as a giant text box where you'd have to type out commands to buy milk on your phone (with curl). That wasn't because it was impossible, it was because (1) most people can't write code, and (2) even if you could, typing out a command for each item would take far longer than just tapping a button. So they did what every shopping app does: they built a screen with a search bar and buttons for adding and removing things from your cart.
What if though, you not only knew how to program, but also knew how to program as well as an experienced developer and could do it faster than any developer on Earth, all without having to pay the large salary such a person would command, if they could even exist. Well if this were true, then you would be in our current reality. And in our current reality the assumptions that drove Steve Jobs toward touch screens and fingers no longer hold.
Until now, the vast majority of computers have allowed us to do things we find valuable, by showing a predefined set of actions with which we have the power to compose in a specific order to attain a valuable thing. In a way each button is like a piano key, but you still have to play them correctly to make music. For example, here are the predefined set of actions i can see in google docs right now
Here are the predefined set of actions you can take in the woolworths app when searching for toilet paper:
And here are the predefined set of actions in SPSS
We’ve gotten so used to expressing our desires by composing button clicks that it’s easy to conflate the button itself as thing we desire. When really, they are just how we have converged on representing the actions you can take. They are not the only and as of recently not even the best way to achieve our actual ends which are buying milk or a making a graph.
Interacting with computers mediated by buttons makes sense only if the person looking at the screen is the only thing that can decide how to compose actions to accomplish exactly what they want. For the longest time this was true, today though, this is no longer true because for most things, sonnet 4.6 can at the same time select the correct actions and express those actions as code, faster than you can express them with a mouse or your finger. The only problem we have then is how to expose actions that were previously buttons, to AIs. Weirdly we already have. What we need exists (metaphorically) under the button. It’s the code human developers have spent all their time writing for the last 50 years. As of today, most of this code is only designed to be reached by humans clicking buttons. The actual code that runs when you click the button, exists though, and If an AI had access to it, it could take actions on behalf of a person much faster. Mostly, AIs don’t yet have the ability to interoperate with the code under buttons, so they can neither see the actions nor execute the actions an app can perform, even though it is intelligent enough to both understand what we want and reify it using those actions.
If the problem isn’t AI intelligence then do we just have to find a way for AIs to be able to communicate with the programs we care about on our behalf? Basically yes, we once thought the answer to this was something called MCP, but this doesn't seem to be the case anymore so we’re not gonna talk about it. Instead it's command line interfaces which are proving to be the best way for AIs to do things on your behalf. In an ironic twist of fate, the mouses and windows Steve Jobs borrowed from Xerox to replace the command line are now themselves being replaced by command line. Command line applications are just programs that you interact with via text. For example here’s a command line application called yt-dlp that lets you download youtube videos:
If you type:
It will download the video for you. Easy. Turns out, modern AI’s are really good at writing commands like this to achieve things on your behalf because it’s just text.
For the last few months it's been clear to us that AIs are exceptionally good at programming. You’re probably imagining this means an AI making stuff like websites or apps that people would then use. Wall street certainly thought this a few months ago, to the extent that many companies which only make money by selling a single website or app, suddenly became less valuable. It is true that AIs are exceptionally good at building traditional software like this, but they’re equally good at a category of programming we don’t have a name for because it doesn’t fit with our fundamental assumptions about what programming is and should be for. This kind of programming is one where the program itself has no value, only the results it generates do. They are programs which are entirely customized to your very specific task and deleted instantly after they’ve been run. For example, you ask claude if your psychiatrist emailed you, claude writes a full Python program to search gmail. It then executes it, gets the result and notices it only got back 30 emails. So it writes another full Python program, this time adjusting the number from 30 to 1000 and based on that result replies “yes.” From your perspective this took 10 seconds and all you saw were the words “thinking…”
These are disposable programs. The point of these programs is just the result. It’s kind of like when you’re using a calculator to do your taxes, you input some numbers, get a result which you write down and move on. Once the AI has the result, it doesn't matter what happens to the program and something that would have taken a human days to write will be executed then deleted with the same care as an accountant pressing AC on a calculator. We were born with programs being something complicated to create, something that is impressive when done well. When we say AIs go beyond the power of human programmers, not only are they superior at the normal type of programming we’ve grown up with. They are also superior at programming in ways we didn’t know you could program.
Today, there are probably some problems only a team of cracked developers can solve. Perhaps Opus 4.6 is worse at coding than such a team. But for everything else Opus 4.6 will do the exact same quality of work, in seconds over and over again for the entire night. This change in quantity is also a change in quality. Developers paying hundreds of dollars for a Claude Code subscription which they then use to make a product for a SaaS company or add features to one is what strapping an engine to a carriage looks like. If you have access to something that knows how to program as well as an experienced developer, why do we need the saas app or feature anyway?
Wall street worried that companies would vibe code their own version of trello and cancel their subscriptions, what would be worse is companies not needing to vibe code trello at all, because agents can manage their own tasks better with python and tell you what they’re working on in English.
When GPT 5 came out, I got it to do my mom's shopping. She could send a picture of a handwritten list to a whatsapp number, GPT would search then add groceries to her cart using disposable scripts but I stopped working on it, it had some bugs and I got demotivated when ycombinator rejected it.
Whenever I’ve seen my mom shopping since then, I’ve felt deep guilt because I didn’t make this fully usable. I felt even more guilty when sonnet and opus 4.6 came out because I knew that they would absolutely nail this task even better than GPT 5. Part of me hoped or assumed that someone would do this for me when openclaw got big. But no one ever did. Part of me hoped or assumed the companies themselves would do it, but checkers made bread tinder. It's become clear to me that stupid minutes won't go away on their own. But they will go away the when we decide they should. A tiny amount of time spent opening doors will yield huge returns in our every day lives, because the same intelligence currently building apps in one shot can more than easily do our sludgework, we just have to let it.
I don't know how much we should use AIs for creating art or writing or how we should aesthetically value what they do. I do know it's pointless to argue about this when there's a million things we obviously should be using AIs for that we aren't. And I do know that it's better for humans to spend more time painting and less time comparing toilet paper sales. So that's where I'm starting.
Pelicart Details
Pelicart will be free in beta. Once it seems to be working well with just Woolies Dash I'll invite more people to the beta, probably start charging something and eventually add Sixty60 and PnP and Dis-Chem, so you can genuinely and completely never have to spend stupid minutes on shopping again. Pelicart isn't designed to replace browsing through stores. A couple of days ago I was hungry and tried using Pelicart to buy me some snacks. It sucked at that. I found it way better to just scroll through the app.
Pelicart is also just step 1 for me. I mentioned some other types of stupid minutes earlier like two factor auth, which I think are just as stupid and which I think are just as easy to do away with and which I will do what I can to help do away with.
While writing this I got an email from the read it later app Matter which looked like this and I think it perfectly sums up the direction I see computing going:
Then a few days later i got this
This is what the end of the app looks like.