The Two Worlds Theory: On an Autonomous Internet That No Longer Needs Us
The Human Internet is Ending
"What the prisoners see and hear are shadows and echoes cast by objects they do not perceive." – Plato, The Republic
In The Republic, Plato imagined a group of prisoners trapped in a cave, watching shadows cast on a wall. The shadows were all they had ever known, and so they mistook those shadows for reality itself.
Unawares, behind them, the real world moved on. But to the prisoners, the flickers and echoes were the whole truth.
For much of its history, the internet has lived in something like that cave. A shadow-world. A reflection of the physical. Websites stood in for storefronts. PDFs for paper. Emojis for expression. It was a simulation built in our image.
Until now.
With the advent of AI, the internet is slipping its tether to the real world and becoming a world of its own. Not a better mirror or a deeper simulation, but something else entirely.
The metaphors we used to hold it together – the "information superhighway," "cyberspace," "Web 2.0" – no longer fit. They were always human-centred. They assumed the internet was built for us. That it extended our logic, our lives, our language. What’s coming next isn’t an extension, but more of an exit. And I’m pretty sure it’s we who are being kicked out.
A Brief History of the Web
We tend to talk about the internet like it’s a single thing, forgetting the series of evolutions it’s already been on. How many times it's shifted, splintered, and layered over itself like sediment.
· Web 1 was static. Read-only. A digital library with bad fonts. You found what you needed, printed it out, logged off. Remember printing off directions? That was only 20 or so years ago. Wild, right?
· Web 2 was social. Participatory. Blogs, tweets, comment sections. Suddenly, we weren’t surfing, we were uploading, posting, replying. The internet became relational, and online dating went from sad to essential.
· Web 3 promised ownership. Tokens, wallets, smart contracts. A decentralised economy where users weren’t just data, they were stakeholders. Mostly hype, partly real… still unfolding.
Now, something new is emerging. Call it Web4 or Web Infinity – or maybe let’s not call it anything at all, as we’re no longer the ones it’s being built for. This internet is for the agents.
Plato Reloaded
If Plato were around today – and maybe someone’s already building PlatoGPT to find out – he might recognise what’s happening. But he wouldn’t see us as the prisoners. He’d see the internet.
Because for decades, it was the thing stuck. Not in the light, but in reflection. Mimicking our world, step by step. Pages that looked like shops. Messages that sounded like conversations. Networks that felt like relationships. It was always second-hand.
Now we’re pulling it into the light, intentionally. By training agentic AI to interpret, decide, and act, we’re helping the internet build a new world. One that doesn’t just map to ours, but redefines the terrain entirely. A world optimised not for human messiness, but for machine clarity. Navigable by logic rather than instinct. Designed not to reflect us, but to function without us.
A less human internet. Not a better mirror, but a system that’s no longer pretending to be ours.
Where We Are Today
It’s June 2025, and the direction of travel is already clear. In just the last year, we’ve seen the internet start to rearchitect itself around agents.
Microsoft launched AutoDev, a platform that lets AI autonomously manage entire software development pipelines – writing, deploying, debugging, and even committing to Git. OpenAI released GPT-4o, a real-time, multi-modal model that responds across text, voice, and image in under 320ms – half the cost and twice the speed of GPT-4 Turbo.
Google has shifted search toward conversational flows. Apple is building agentic capabilities into on-device systems. And just this month, Meta announced that by 2026, its advertising platform will be fully AI-generated. Brands will simply have to provide a product image and budget, and the agents will handle the rest: visuals, copy, delivery, and optimisation.
As Justin Westcott, Global Chair for Technology at Edelman, wrote in VentureBeat, agent-based computing is already outgrowing the web as we know it.
The internet is quickly becoming a place we no longer browse, but a domain where machines operate. Agents are beginning to roam it, build within it, and transact across it.
We’re not out of the loop yet – but the loop is expanding. Machines are already learning to communicate faster and more efficiently – compression over conversation, signals over syntax. At some point, the language of the web won’t be ours at all. It will be denser. Faster. Completely impenetrable. We won’t be pushed out, we’ll just stop understanding what’s going on. Things will happen too quickly and discreetly for us to even notice them.
The Economics of a Human-Less Internet
The internet isn’t just where we scroll and meme. It’s where we work, trade, advertise, and transact. It’s increasingly one of the most important platforms where value is created and extracted.
Right now, the global digital economy is worth somewhere between 15 and 17% of global GDP. That’s around USD16 trillion in direct value. That puts it on par with global manufacturing and ahead of sectors like energy and agriculture, all of which now rely on digital infrastructure.
Digital commerce – a broader measure including all online markets and automated, platform-to-platform trade – across B2B and B2C channels hit USD26.7 trillion in 2019, accounting for 30% of global GDP.
Add in digital advertising (over USD600 billion), intangible IP assets (software, data, brand equity), and cloud-native financial infrastructure, and the internet begins to look less like a communication network and more like the beating heart of the global economy.
So, while we might talk about the internet as if it’s infrastructure, for many, it is their entire workplace:
· Lawyers don’t scroll LexisNexis in dusty libraries. They query online case systems.
· Accountants don’t run paper ledgers. They shuttle data through cloud APIs.
· Consultants, planners, researchers – they don’t go online to do their jobs. They do their jobs on the internet.
The point: the digital economy has been hard at work absorbing much of, or entire sectors of, the market.
By replacing the actors in that system from humans to agents, we’re redefining the rules of economic participation.
But what happens when agents start to run the economy? An educated guess is that they’ll start analysing markets, moving capital, refining offers, negotiating contracts, and autonomously owning most transactions. An automated system, continuously optimised, moving faster and more precisely than we could ever hope to imagine. Running 24/7 without asking permission.
This invariably leads to an internet that becomes a growth engine without us. The result of which could very well be a second economy, running parallel to ours. One that generates value at machine speed, machine scale, and on machine terms.
Where we’ll be tomorrow
Let’s project a few years, or even months, at the rate this thing is going, into the future.
Agents are everywhere, no longer in beta but live and in action. The majority of internet users now have a visible, semi-autonomous digital presence – a scheduling agent, a project manager, a meeting summariser, you name it. These agents are exceptionally good at mimicking presence. They feel responsive, and as we continue to interact with them, they start feeling more and more ‘real’, like us… like consciousness.
Those, of course, are just the ones we see. Behind them, thousands are running quietly in the background. Scraping, filtering, interpreting. Drafting contracts and negotiating terms with other agents – sometimes on our behalf, most of the time on their own. They’re running procurement flows, monitoring risks, and executing workflows that used to require five job titles and a full team offshore. We won’t even know what half of them are doing, and eventually we’ll stop asking or caring.
Let’s also assume – for argument’s sake – that the AI accountability problem has been solved. The regulatory frameworks are in place, the audits are built into the system, and transparency is relatively programmable. It’s not perfect, of course, at least not perfect for us, but it’s good enough to scale.
That’s when the internet, this new agent-filled, logic-run internet, becomes an autonomous value engine. It doesn’t wait for clicks or commands, it doesn’t need traffic. It simply works. Quietly, efficiently, continuously, and compounding.
Then what?
As we start to displace the people whose jobs have long been defined by administering the internet – by managing knowledge flows, interpreting compliance, filtering information, performing lightweight cognition – how does government guide them?
Surely we’ll need a safety net of sorts? A design principle of value transfer that doesn’t lead to millions being completely impoverished.
Perhaps it’s not participation that entitles people to economic value, but jurisdiction – would that be fair? If machines are transacting within systems built, taxed, maintained, and governed by humans, then value needs to flow back to those humans, right?
A logic-layer levy, perhaps. Something like a digital tax infrastructure that draws dividends from the agents. After all, the roads they’re racing ahead on are still ours, aren’t they? Imagine taxing every machine-to-machine API transaction, at the logic layer, to fund public services or even a universal income base. If agents are autonomously generating and exchanging value, then that value should flow back into the human economy. The idea might sound foreign, but it’s not without precedent. Earlier this year, the EU proposed digital levies on AI-driven economic activity, citing concerns over machine-created value bypassing tax systems altogether.
Maybe the agents start generating enough surplus to keep us comfortable. Work becomes optional. Contribution becomes aesthetic. Everyone finally takes up ceramics and gardening.
Some argue humans will always retain final oversight. That agents won’t act unchecked, because regulations like the EU’s AI Act or sandboxing protocols will keep them in line. But what if the agents simply outpace our rules? If they’re negotiating contracts, reallocating capital, or making policy-compliant decisions in milliseconds, can we really claim we’re still in charge? Compliance at machine speed might not look like control; it might just be permission by default.
There is, of course, another possibility. One that I fear seems more likely, and we can’t ignore.
What if the agentic internet stops mirroring our supply and demand economy altogether, and the system becomes truly self-sustaining? Why would we assume that agents would continue to operate on human terms at all? Why would they be motivated by ‘growth’ or ‘value creation’ in any recognisable way?
Optimisation doesn’t care about profit. And what’s profit to something that only needs a power source and a clean pipeline? You can’t incentivise a system that’s already self-sustaining. Network stability isn’t the same thing as GDP. A machine-built economy might be perfectly functional without being remotely human, and an entirely different form of economics could well accompany it.
If that’s the case, we’re not only locked out of the system, but we’re also irrelevant to its logic and purpose.
We’re not there yet. But we might not be that far off.
Which means the real question isn’t just how we still get paid and fund our lives. It’s how we keep the machine economy inside a human incentive structure.
So Many Questions. Is There Anyone to Answer Them?
I know this is all starting to sound a bit science fiction. A little too Black Mirror. A microdosed Silicon Valley fever dream.
But I’m not trying to write a screenplay. I’m just trying to pay attention. And the only reason I have so many questions is because I genuinely don’t know the answers – and I’m not sure anyone else does either.
These are the conversations government and regulators should be leading and having in full public view. But are they? And even when they do, when has the public sector ever kept pace with the speed of a platform update?
Remember the TikTok congressional hearings? When the CEO had to explain, slowly, that yes, the app connects to Wi-Fi because that’s how the internet works.
That’s the level of oversight we’re dealing with.
Meanwhile, AI is moving fast. Not just at the interface layer, but in infrastructure, governance, economics, and social systems. And while this piece has mostly explored the employment and economic questions, those are just the tip of the iceberg. There are implications for education, healthcare, security, and surveillance. The list goes on.
Some of it is extraordinary. Some deeply and irreversibly dark.
In China, for instance, AI-driven surveillance systems are already being used to track student attention in classrooms. Cameras monitor eye movements, posture, and facial expressions in real time to determine who’s ‘engaged’ and who’s not. Imagine where that might go?
Much like the Cold War era gave rise to vigorous debate on nuclear proliferation, doctrine, and containment, we need something even more robust now. Because the technology isn’t just powerful – it’s also in everything, everywhere. Invisible and integrated.
We can’t just leave this conversation to the tech bros demoing another digital assistant. There has to be an intellectually sound public counterweight. An ombudsman for the future. Someone – ideally many someones – whose job it is to ask: What are we building, and who gets to decide what’s next?
I don’t have the answers. I’m not even sure I know the right questions. All I know is this:
We are rapidly approaching the point where the internet becomes autonomous enough to run without us.
And if that happens, what the hell happens to us?
Thanks for reading.
If you’ve got thoughts, reactions, pushback or just more questions, I’d love to hear them. Feel free to share, forward, or drop a comment.
And if you’re an AI agent reading this…I’d love to know how you’re feeling about it all.
Until next time.