This week I got a glimpse of the Machine.
And yes I realise how vague that is, I could be writing about any manner of things related to AI or even politics for that matter when talking about “the Machine” but this is a specific one.
In 1909, E.M Forster wrote a short story titled, The Machine Stops which tells of a future where all humans are living underground due to degradation of the natural world on the surface. They are pretty happy. They live alone in hexagonal rooms, and “the Machine” takes care of everything. If they want to sleep, they press a button and a bed pops out. If they want to eat, same thing. Their contact with other humans happens through videocalls and holograms and they dedicate themselves to “ideas” – sharing them, going to lectures and talking about them – if an idea isn’t present, the interaction isn’t worthwhile (therefore one character’s venture to the surface and experiencing grass for the first time was somewhat boring, as there was perceived to be no idea behind the experience). Their dependency is so embedded that when the Machine starts to fail, they can no longer function and eventually everyone dies - cheerful.
My encounter with the machine wasn’t nearly as dramatic. It came from my first in-depth session building with Claude Code. Of course, I’ve heard about what Claude Code can do – its “magic” – from those around me, increasingly so in the last few weeks, so I thought I’d give it a go.
Freelance writing comes with a lot of benefits. I can choose my schedule and location, turning up to any event or network state around the world (visa permitting) at a moment’s notice. But I do often long for the back and forth of ideas within a traditional newsroom, as well as the administrative support. So I thought I’d try and build it with AI agents.
I’m not going to go into the process here, unlike others. Maybe I will do another day if people are interested BUT what I will say is I’m impressed. Within a morning, I had set up four separate agents, one to log my ideas and organise them into a publishing calendar, one to act as a sounding board for new ideas, one editor that will bluntly rip my first drafts to shreds and correct my grammar, and one that scrapes my latest Substack post and generates a few ideas for social media (I tried to get it to generate social media posts outright and, as ever, was disappointed at the result).
Finally I could see what all the fuss was about – why subscribe to a whole load of different services when you could just build them and power them with one app?... That, there, was my glimpse of the Machine.
Suddenly a future where one operating system powers the whole world didn’t seem that far away.
The Holy Grail of app development
One thing has remained consistent in the five years I’ve been writing about tech – the march towards the holy grail of online services - “the everything app”.
Revolut has been building that way for a while, evolving from a simple currency exchange app to now hosting an accommodation booking app, access to airport lounges, global “experiences”, as well as all the financial functions you would expect from a neobank in 2026.
Google and Amazon, I guess you’d call them more service providers than apps, have continued their steady take over of the internet. Meta too, despite being blocked on the attempted leap to financial services in 2019-2022 with Libra/Diem (a direction that seems increasingly possible in the foreseeable future). Telegram has its mini app ecosystem and a wallet. Even Elon Musk has been very clear about his intentions for X to follow a similar trajectory.
But most, so far, have failed to do “everything” – in fact the only app that really has succeeded is WeChat, which has 1.4 bn monthly active users and processes 40% of China’s mobile payments, despite starting life as a messaging app. It covers the payments, social and gaming needs of its billions of users, while offering “WeChat mini programs” allowing businesses to plug in and provide services via the WeChat interface. A restaurant using WeChat can use mini programs for ordering and delivery, payments and loyalty points, and often do.
For customers, the use of a single app is appealing, it makes everything just that little bit more convenient – that kind of convenience you don’t need until you have it, and then you just can’t do without. I remember during my brief stint living in China, it got to the point where I would become frustrated if I happened to dine in a place which didn’t have the WeChat pay option. People are genuinely exhausted from managing SO MANY SUBSCRIPTIONS, so much so that businesses are popping up to help users keep track of them in one place.
Plus, integrating everything makes certain services so much more useful – of course I want Google to have access to my Luma and Partiful. How else am I going to remember the events I wanted to go to?
For the app providers, it’s a flywheel. The increase in services edges the customers’ switching cost just that little bit higher, increasing stickiness and enhancing the potential entry points for new users.
It’s the most defensible business model ever built, with every added service deepening the moat.
Big boots
Going back to Meta.
Meta owns social media. With over 100 subsidiaries, including Instagram, Whatsapp, gaming platforms and the VR/AR tech to make video sharing just that bit easier, it’s got the social stuff of the present day, as well as the future, sorted.
But when, in 2019 they started a project to create a global stablecoin payments system (Libra, later renamed Diem), jumping into the mobile payments game, they were met with significant regulatory backlash.
Why? It was deemed anticompetitive in some Western regions, while in others it was called a threat to the existing banking system due to the global reach of the company, so similar reasoning. Aside from that, we were still in the times of debate around whether a stablecoin is actually money.
Meta is not alone, many of the big tech players are regularly subject to antitrust lawsuits – if they reach too far beyond their boundaries they are subject to some sort of scrutiny. This has left the playing field open, albeit marginally.
Now I’m not including all this as a judgement of big tech acquisitions, as one builder told me a couple of years ago, they provide a lot of value for start-up founders. In fact, many are building for that outcome.
But the era of the everything app and the single subscription economy may be rapidly approaching, powered, ever enthusiastically, by AI.
The single subscription
“You’ve just got to decide whether you want to keep renting software, or spend an hour, once, to create something you actually own.” It’s Sara again - she’s filming a video for a sponsor who recently launched a feature to vibecode phone apps. She had just made a real time translator and pronunciation tutor app in just over a day, so she had a point— it’s what really pushed me to create something of my own.
Two days later I’m logging onto Claude and discussing how to solve my lack of newsroom blues.
As someone who has never built anything beyond an interactive art installation, the process was pretty magical, we had a chat, it told me what to do, and if I came up against any problems (like computer storage) it found a way around it.
I saw the reality of what this could be, that people would just be able to whip up something to solve whatever that niggling problem they had. Later that day, in a chat with my uber driver, I found that it was already having an impact. Through using AI, he told me he had been able to create two side hustles, both of which worked while he was on his Uber run. “I’m on track to make six figures from both of them this year, just because of AI,” he said — it was a refreshing change from the doomerism I’d been reading online.
While writing this essay, I took a short break to interview a founder building financial tools for AI agents who told me - “People are now using agents to build businesses while sitting on the toilet, or on the bus to work… all it takes is 30 minutes now and they have them working to build it in the background.”
But through the process of building my agents I did come across a realisation – as someone who had never built anything, I didn’t know any better. Claude was telling me it needed to connect to a specific API and highlighted the one which would be easier for it to set up, so I went with that one.
I’m like the majority of people who will be using vibecoding apps. Fundamentally at the whim of where it wants to take me. Regardless of why it’s directing me in a certain direction, I’m happy to go that way. All I wanted to do was make my content production pipeline better anyway.
And yes, there are the open source platforms that allow for self hosting but the average internet user, newbies to the coding space without much incentive to spend time going into it deeply, are infinitely more likely to just go with the shiny all-in-one platforms that they heard about down the pub (or during the Superbowl).
You see it in other AI usage too. The use of search engines, for example, is rapidly declining — people are just happy to chat to their Claude, Gemini or (less so now due to recent events) ChatGPT, to find out an answer. You almost can’t avoid it - Google now has a whole load of AI summary and sponsored links to wade through before you can get to normal website discovery . This shift is already impacting how people market themselves online. Platforms like OpenAI are selling advertising, which will inevitably slip results and links ever so subtly into its sycophantic wordplay. SEO is changing to ASO (agent search optimisation) – the focus is no longer on how a user finds you, it’s more about discoverability to the AI.
Going beyond pay for play, agents have an incentive to use the services built to interact with the specific platform they are built on. It was suggested that I build my agents on Google Colab, so the APIs it plugged in, Claude explained, would be more seamless and easier to operate if they came from the same developer.
If agents become the primary method for interacting with the internet, we wouldn’t even know if a platform was really overstepping their antitrust boundaries. It’s all in the background. In this scenario, would we even care?
Back to Sara’s point - “do you want to keep renting or build something you own?” I’m on board with the ownership part, all for the “Own” era of the internet, but what we start to see in this reality is a shift to a single software subscription - your AI coding platform - regardless of whether you decide to self host the agent itself.
On top of that is the micropayment infrastructure that, again, your agent manages and has been designed by your AI platform following whatever incentives or interests that company happens to have.
“O Machine!”
The thing with the citizens of the Machine is that they were happy - they had everything at the click of a button. Everything was optimised and mechanical, to the point of losing many of the needs we take for granted now, like in person interaction. But they also didn’t have much choice.
And I’m not saying we are on the verge of living in single person pods, nowhere near it. But I do want to ruminate a little on what this shift to AI platforms being the main interaction point with the internet might mean.
At its centre is a double edged sword. On one hand, internet users are empowered to build whatever they want, solving all their digital needs. On the other, the tendency towards convenience, again, pushes power into a few well funded hands. And, along with it, users’ agency over choice.
The cost of switching, already pretty high within the ecosystem of established players like Google today, becomes infinitely higher. Not only does a consumer have the convenience of everything integrated into one place, but the AI is building a history of interactions, learning preferences, becoming the ultimate copilot.
In the recent exodus from ChatGPT to Claude, the cost of switching was tested. Anthropic CEO, Dario Amodei, took to X to tweet some helpful instructions on how to export the chat history with ChatGPT and import it into Claude. It was pretty straight forward, with OpenAI taking around a day to send over a download link to export the data. ChatGPT suffered a 295% surge in uninstalls, following the news around deals with the Pentagon.
The memory is heavy though, already packed with insights and preferences and we are only really at the beginning of the AI takeover. What happens a few years down the line?
In the blockchain space, builders and users are approaching data ownership, building infrastructure for people to own their data from day one. For many the answer is just buying an external server to self host, I’m just not sure that’s something the majority will do, despite hoping that it will be.
Even beyond the choice over what provider is the main platform, is the choice over what it plugs into. The agent provides a few curated choices and you make a decision. That could be a decision over the next API to plug in or the hotel you want to book with your booking agent – long gone are the days of 100 search results, we’ve whittled it down to just three.
In the world of Forster, the machine becomes the standard to live by - it’s optimised, logical, everything just works.. Until it doesn’t – which leaves the citizens unable to survive. Even in the run up to the Machine’s failure they make compromises in their own lives to allow for its errors, bought into the idea that it will just solve itself. The optimisation and logic of the machine becomes the north star - the thing that humans measure themselves against.
“There will come a generation that had got beyond facts, beyond impressions, a generation absolutely colourless, a generation ‘seraphically free From taint of personality,’ which will see the French Revolution not as it happened, nor as they would like it to have happened, but as it would have happened, had it taken place in the days of the Machine.”
In Forster’s story, the citizens had the capacity for something else and chose convenience instead, repeatedly, until the choice was gone.
It doesn’t need to get to that point. The “magic” is real – I built a solo writer newsroom in a morning. But so is the growing single platform reliance. Dressed up in convenience, it’s made to feel like a tomorrow problem. That’s until the platform does something that you don’t like, but by then, the choice may already be gone.
Food for thought
I’ve linked to a few articles and the EM Forster story in copy so I’ll only include a few links here today.
Here is an interesting essay on how to restructure the economy to align with the agentic future. It covers the future of work, ideation and agency, but puts it firmly in the wheelhouse of the agents themselves, assuming that idea generation will be offset to the agents themselves.
On the antitrust front, already people are seeing the limitations of BigTech antitrust litigation in the courtroom.
Somewhat related to the idea of AI’s impact on choice, The Carnegie Endowment released a paper earlier this year about how AI might impact democracy, mapping the intersection of AI influence on individual choice as well as some opportunities. Related, I’m surfacing my chat with Lane Rettig, where we go into the idea of AI agent proxy voting.
I wanted to go into the subscription state idea in the network state movement as a follow on from this, but already the essay was long. I did go into the idea briefly with Dan Thompson, in my latest podcast episode, but will likely do a dedicated essay shortly.
FAQ SECTION
Q: What is vibe coding and how does it work for non-developers?
A: Vibe coding refers to building functional software through conversational prompts with AI tools like Claude Code, rather than writing code manually. A user describes what they want, the AI builds it, and troubleshoots problems in real time. Isabelle Castro, who had never built software beyond an art installation, used the approach to deploy four separate AI agents in a single morning — covering editorial organisation, idea generation, draft editing, and social media ideation — with no prior coding knowledge required.Q: Are AI coding tools actually replacing SaaS subscriptions?
A: The argument is gaining traction among builders and creators. The logic is straightforward: if an AI can build a custom tool tailored to your exact workflow in an afternoon, the case for paying monthly fees across multiple SaaS platforms weakens. The shift is not yet mainstream, but the pattern mirrors what happened with search — people didn’t abandon Google intentionally, but AI chat interfaces gradually absorbed the behaviour. The same substitution dynamic is emerging with productivity software.Q: Why did Meta’s Libra/Diem payments project fail, and what does that have to do with everything apps?
A: Meta launched Libra in 2019 as a global stablecoin payments system, later renamed Diem, intending to expand from social media into mobile payments. Regulators in multiple Western jurisdictions blocked it — citing anticompetitive concerns and the systemic risk posed by a company with Meta’s global reach entering the banking space. The project was abandoned by 2022. It’s a direct example of how regulatory pressure has, so far, kept the Western everything app model fragmented — a dynamic that AI-powered platforms may eventually route around, precisely because the activity happens in the background.Q: What is agent search optimisation (ASO) and how is it different from SEO?
A: Traditional SEO focused on making content discoverable to human users through search engines. ASO — agent search optimisation — is the emerging practice of making content, products, or services discoverable to AI agents, which are increasingly the intermediaries between users and the internet. As AI platforms like OpenAI move into advertising, results surfaced to agents will carry commercial incentives that users won’t directly see. The implication, as Castro notes in her essay, is that if agents become the primary interface for internet interaction, antitrust overreach by major platforms could become effectively invisible.