A Startup That Builds Software From Plain English Just Raised $400M at $13.3B. The Building Is Getting Cheap. The Deciding Is Not.

On 12 August, Lovable, a Stockholm startup, raised $400 million at a $13.3 billion valuation, double what it was worth in December. Menlo Ventures and an EQT-managed fund co-led, with Tencent and others in. This is not a slide-deck story: Lovable's annual recurring revenue has nearly tripled from $200 million and is tracking toward $600 million by the end of this month. What it sells is deceptively simple. You describe the software you want in plain English, and it builds it.

What vibe coding actually commoditises

Strip away the hype and Lovable is industrialising the part of software that used to be slow and expensive: turning an idea into a working, deployed app. Front end, database, plumbing, the scaffolding a small team once spent weeks on now takes a prompt and minutes. That is genuinely valuable, and the revenue says the market agrees. It also means the scaffolding is no longer where the value lives. When anyone can generate a competent app or dashboard on demand, having built one stops being an advantage.

The thing that does not get cheaper

Here is the distinction worth holding onto. Vibe coding makes it cheap to build software that records, displays and moves data. It does nothing to make the hard decision inside that software any better. Generating an app that lists your overdue invoices is now trivial. Knowing the margin-optimal price to quote, the safety stock that will not tie up cash, or the shortest way to pick an order is not a coding problem at all. It is a data-science problem, and prompts do not solve it. The build is commoditised. The decision is not.

What it means if you run on an ERP

For a business already running on a system like Dynamics, this cuts a clear line. The cheap, fast layer, spinning up a screen, a form, a view, is now genuinely cheap and fast, and you should treat it that way. The expensive, valuable layer is the optimal call the system cannot make on its own: the last mile between what the ERP records and the decision that actually saves money. That is not something you prompt into being. It is a system of intelligence you add on top, built on the same governed stack and pointed at one decision at a time. It is the same lesson underneath cheaper models generally, from capable agents that now run on a single GPU to tiered model pricing: as the raw capability gets cheap, the edge moves to knowing exactly what to point it at.