Yesterday, Lovable confirmed a $400 million Series C round at a $13.3 billion valuation — doubling in eight months. The Stockholm company sells the ability to turn a text description into a working website or app, and co-founder Anton Osika describes the ambition as building “the last piece of software”: the tool that will turn a billion people into creators. Annualized revenue stood at $500 million in June, and 60 million projects have been built on the platform.
It is an impressive shovel in an ongoing gold rush. But there is a question the valuation rests on that is rarely asked outright: will the gold vein stay where the shovels are digging? What happens to companies that sell app-building if the need for apps and websites shrinks — because AI agents become people’s primary, perhaps only, interface to the digital world?
The thesis of the disappearing interface
The argument is not new, but it now has infrastructure. Satya Nadella formulated it back in December 2024: business applications are essentially databases with business logic, and once the logic moves to the agent layer, the application layer “collapses.” Since then, Anthropic’s open protocol MCP has become the most widely adopted standard for how agents use services without going through an interface — industry statistics now point to close to a hundred million SDK downloads per month and thousands of servers. OpenAI’s and Google’s web agents, which click through pages on the user’s behalf, point in the same direction: the user describes a goal, the agent executes it, and no human ever sees the web page where it happened.
If that becomes the default flow, what an app is changes. The interface stops being the product. The service behind it — data, logic, payments — becomes an endpoint that agents call. Whoever pays today to have a polished surface built will pay tomorrow for something else: a well-defined API and a reason for the agent to choose that particular service.
China is ahead
That this is not theory is visible in China, where the move away from standalone apps began nearly a decade ago. WeChat’s mini programs — now numbering in the millions — made standalone apps unnecessary for large parts of the service market; the super apps became the way in. Now the next layer is being added: since June, Tencent has been testing the assistant Xiaowei, which completes tasks by using the mini programs itself. Alibaba’s Qwen assistant reached 300 million monthly users across Taobao, Tmall and Alipay early this year, and ByteDance has upgraded Doubao so that it can book tickets on its own through Douyin’s e-commerce. Chinese consumers never needed to leave the super app — increasingly, they no longer even need to open the mini program. China is not an anecdote here but a leading indicator: the pattern suggests that the final step, to the agent as the way in, happens fastest where distribution is already centralized.
The counterarguments
Three objections deserve to be taken seriously. The first concerns the customer relationship: brands will not voluntarily reduce themselves to anonymous endpoints inside someone else’s agent. The protocols now being built for agentic commerce — such as OpenAI’s and Stripe’s Agentic Commerce Protocol — are explicitly designed so that the seller remains the merchant of record and keeps the customer relationship. The battle over who owns the customer — the agent platform or the service — is not settled, and it will become regulated terrain: the same gatekeeper questions the EU asked about app stores are now returning about agent platforms.
The second objection is that not everything is an errand. People happily delegate booking, comparing and ordering — but entertainment, games, social life and brand experiences are surfaces people want to spend time on, not tasks they want to be rid of. Interfaces will likely disappear where they are friction, and survive where they are the point.
The third is that apps may not disappear but change roles: from surfaces for humans to backends for agents. Someone still has to build the service, the data model and the logic — and tools of Lovable’s kind already build more than interfaces. Add what might be called personal software: when building your own tool costs an evening of prompting, it no longer matters that nobody else will use it. That market grows as agents get better, not despite it.
The pace
Platform shifts take longer than the rhetoric suggests. It took seven years from the iPhone launch until Americans spent more time in apps than on mobile and desktop web combined — and the web still exists. My base assessment is therefore that the agent as entry point dominates errand-heavy consumer flows — travel, shopping, government services — within three to five years, while B2B and experience surfaces move more slowly.
But there is a faster scenario, and it deserves to be marked clearly. Every brake that made the mobile shift a seven-year journey is missing this time. No hardware shift is required — nobody has to buy a new phone. No new distribution has to be built — the assistants meant to become the way in already have hundreds of millions of users. And the agent protocols mature at software speed, in months rather than years: MCP went from launch to industry standard in under two years. China shows what happens when the brakes are gone: Alibaba did not have to convince anyone to install anything new — the assistant was placed inside apps people already used, and counted 300 million monthly users by early 2026. Add the pattern that has marked the whole AI trajectory: in exponential developments, forecasters systematically underestimate the pace, and the projections of recent years have been revised again and again in the same direction — toward shorter timelines. In the fast scenario, it is not three to five years but perhaps eighteen months before demand for newly built consumer surfaces starts to fall. For companies that live off surfaces being built, that is the difference between having time to reposition and not. It is a scenario, not a forecast — but anyone building a business plan on the base assessment should stress-test it against the faster path.
None of this means Lovable’s market disappears this autumn. It means the value of what the company sells shifts: from making interfaces cheap to making services buildable. The irony is that the same models that make Lovable’s product possible are also the ones that could make its original deliverable — the surface — increasingly dispensable.
For European SaaS and no-code companies, the conclusion is concrete. Build API-first and treat the interface as replaceable; own the data, the logic and the customer relationship, because that is where the bargaining power against the agent platforms will sit. And for the Nordics there is a larger point: Lovable was built in Stockholm in three years and is now valued at $13.3 billion. If the entrance to the digital world is changing shape, it is better for Nordic companies to help build the next entrance than to optimize for the one on its way out.
Sources: TechCrunch on the Series C · Lovable, Series C · Forbes on Lovable · Nadella on the B2G podcast (via Windows Central) · The MCP specification, July 2026 · MCP adoption statistics (Digital Applied) · Bloomberg on Xiaowei · CNBC on agentic commerce in China · Richard MacManus, The Agentic Web · Fast Company on the customer relationship · Agentic Commerce Protocol · commercetools on ACP · TechCrunch 2016 on the app ecosystem