Forget The Metaverse, The Future Web Is A Marketplace Of Tools And Capabilities

Technology publications and mainstream media have spilled a lot of virtual ink over the death of the Metaverse this week, and then even more over Meta pulling a volte-face and declaring that it was still hangin’ on in there. There are interesting stories still to be written about what went wrong with the Metaverse as a concept and as a practical reality, but The Interline would argue we captured the first strokes of those three years ago, and insufficient time has passed to do it again.

Nevertheless, it is remarkable news that Meta claimed this week to be shutting down its flagship Horizon Worlds platform – or at least making it unavailable for VR headset users, and transitioning it to a flatscreen experience only. Which, well, sure.

As a reminder, Horizon Worlds was pegged as being the do-it-all, first party vision of what Mark Zuckerberg referred to as the “embodied internet” – a single (read: heavily instanced) digital space where people would go to live, work, and have both everyday experiences and remarkable adventures, all without leaving the house.

Horizon Worlds was also, lest we forget, a place you could buy your dead-eyed avatar Balenciaga, Thom Browne, and Prada outfits.

Just like going back and reading articles like the one above, it is, frankly, a dissociative experience to rewatch the 2021 Facebook Connect video where that “embodied internet” phrase first entered the popular parlance. As an artifact of the COVID era, when every executive seemed to lose their collective sense of time, and assumed we’d all live in sweatpants forever, it’s a remarkable time capsule. But it’s also evidence of just how quickly time can erode ideas that were pretty wonky to begin with.

There were a lot of components to the Metaverse vision, some of which The Interline still sees as valid, even though they manifest in videogame walled gardens. And proponents, including Meta’s Andrew “Boz” Bosworth in his announcement that Horizon Worlds isn’t actually shutting down, will tell you that the idea is “misunderstood,” that VR was only ever an ancillary part of a much deeper idea of taking part in a shared digital using your phone, now, apparently constitutes being in the Metaverse? Again: sure.

However you slice it, though, that hazy pandemic-era period was built on a faulty assumption about the web: the idea that people want to get into it and walk around, rather than treating it as what it is now, which is an application distribution environment for most consumer and enterprise software except videogames.

Perhaps the most frenzied instantiation of that idea was the temporary craze for small nations opening up virtual embassies where people could apply for visas and suchlike. The last five years have absolutely not been kind to that as a concept.

But as silly as the idea of putting on a headset to take part in embodied bureaucracy sounds, the pendulum swing from that conception of the web, as something you want to be immersed in, is tracing a wild arc in the opposite direction – to the extent that the future of online applications looks more like dry, dispassionate databases, atomised tools, and hyper-granular capabilities that barely need a user interface on top of them at all. 

We’ve previously written, and interviewed brand executives, about the idea of composable software – applications that build themselves on the frontend, from predefined libraries of content and capabilities, and that are also capable of extending their footprints.

It’s not a remotely bold statement to say that we see this as a pretty inevitable eventual outcome of the short-term “build your own software” movement. 

Right now, services like Lovable, Replit, and the updated Google AI Studio all allow anyone to build web-native applications, and beat-phase AI “vibe coding” platforms like Glaze extend that same flexibility to local MacOS applications. People like building their own applications for their own idiosyncratic use cases, and corporations are certainly eyeing up the same possibilities, even though the decision-making matrix and the end state will look pretty different for the enterprise.

In both cases, though, the next logical step is software that assembles itself at runtime. This is distinct from traditional configuration or customisation, where customers (or the consultants whose service days they’ve been incentivised to buy) adjust the look, the navigation, the integrations, or the data hierarchies of the tools they buy licenses for. 

Where do those centrally-developed and fixed-function applications live? The web.

Where would a library (or more likely a marketplace) of fine-grained tools and capabilities that could be opened up to AI agents looking to assemble them live? The very same place that people thought we’d want to run around it, and which now looks much more likely to become a massive repository of databases and tool calls than a 3D playground.

What we haven’t analysed – at least not yet – is the likely diffusion curve for this kind of shift in the conception of web applications, and of the web itself. This is, very probably, going to follow a similar trajectory to the cloud, with SMEs that have very fluid roles jumping in first, and embracing the idea that web applications can be more malleable, less predictable, and inherently less secure. And then enterprise uptake will follow.

To be clear: The Interline does not expect anyone who uses a big two ERP platform right now to vibe code their way to a replacement – even if that replacement could, at least in theory, live in a well-sandboxed environment, have a more modern architecture, be more reflective of real business demand, and be a better citizen of both the internet and the modern brand tech estate than SAP or Oracle.

Big companies may use the likes of Lovable for prototyping functionality that they then ask their locked-in vendors to replicate, but they’re unlikely to then use those tools themselves.

Small companies, though? They absolutely will. Because why would they pay SaaS fees for a platform that fits half their requirements when they could natural-language-prompt their way to something bespoke? That something might be shoddier, unsafe, and have a bunch of other drawbacks, but it would be, for their purposes, feature-complete.

And by the same logic, why would an SME, way down the pecking order on a custom advisory board, wait for a big tech vendor to develop a feature they want, if they could instead ask an AI agent to either build it for them or, more likely, pick it off a virtual shelf of pre-defined tools and capabilities?

Now, this might sound like our own version of the infamous Meta pivot, but in reality it represents a more grounded version of the latest in Zuckerberg’s series of predictions. 

After buying the wildly over-hyped Moltbook (a “social media” platform for AI agents) last week, Meta is now committing to the idea that “every business will have their own AI,” and that those AI agents will interact with one another. 

As positioned, the vision here seems to be that these agents will converse and collaborate to perform higher-order tasks, and while the idea of one agent watching the programmatic ads the other agent created is amusing, The Interline thinks that a much more likely outcome is that one company’s AI representative will ask its opposite what tools, skills, and data it has, and it will use them or acquire them for itself.

This is an implausible-sounding but actually fairly reasonable-to-infer future of web applications – one that disintermediates users and vendors, at least from the technical side of things, and from the end user experience.

And this betrays a very different idea of what the internet is going to be. From people assuming we’d want to immerse ourselves and swim through it just half a decade ago, to a near future where AI agents select skills and tools from libraries, or go out and acquire the modular pieces of software their users want, but sidestep the need for a UI entirely.

We could also go one further: why would an enterprise put together an extensive RFP, solicit demos from multiple vendors, and spend time shortlisting and selecting (and often being sold false promises) if an AI agent could just query a SaaS platform and identify and test the tools it offers?

We obviously don’t expect any big enterprise companies to buy software or modules sight-unseen, based on the recommendations of their AI, but we won’t be surprised if the next stage of identifying and working with web software starts with either MCP tool calls or an LLM spinning up a virtual browser instance and playing around with demonstration software.

These feel like silly things to predict, given how much time we all spend interacting with, typing into, and mousing around interfaces designed for people, in software suites we bought seats for after carefully evaluating the best the market had to offer. But if the next bet on the future of the web is agentic, then we, as humans, really flatter ourselves if we think the limits of AI will be deploying it to use software, rather than making and buying it as well.