Lovable's CTO says enterprises show up with thousands of apps their staff already built

A sales leader at Uber Eats who knew the restaurant pitch process end to end built a tool on Lovable that produced a pitch customized for each restaurant ^[8]. The first version took a few hours or a day, and later changes about 10 minutes ^[9]. Changes like those used to wait months on an engineering backlog ^[10].

The people doing this sit in finance, HR, engineering, product and design ^[11]. The call to Lovable comes later, when a company has thousands of apps and wants to know what to do about them ^[2]. Hedin said companies are also replacing costly internal software that offered little customization, and that users own what they build and can take it and run it themselves ^[23][24].

Martin said revenue was about $200m when he visited the Stockholm company last summer ^[4], roughly a third of the figure Hedin gave on Wednesday, so revenue has about tripled in half a year ^[6]. In August, a $400m round valued Lovable at $13.3bn ^[5]. The report of the session does not break the $600m into company contracts and individual subscriptions ^[7].

Security teams sometimes worry about all the new code, "rightfully so", Hedin said ^[15]. He said AI agents also change security inside large companies ^[16]. "I think before, a lot of people have been able to rely on kind of security by obscurity, where like the permission system internally in a large enterprise has not been perfect, but no one knows how to navigate it. Now you can have an agent navigate it," Hedin said ^[14].

More than a million new projects a week comes to upwards of 50 million a year for Lovable's scanner to cover ^[19]. It fixes critical flaws in dependencies and proposes fixes for the rest, and Hedin said the tool is free for all users and the first in its category ^[18]. Martin raised the pattern of vibe-coded apps that took off and then leaked user data; Hedin said builders have to assume vulnerabilities will exist, and pointed to a flaw recently found in the curl command-line tool ^[20].

Lovable has never offered a model picker, and for a single task it may use more than 100 models, weighed on cost, speed and quality ^[21]. Hedin said customers ask for the routing after the industry's talk of token maxing turned into bills ^[22]. Without a picker, a buyer cannot specify which model handles a given app's data.

The sort that makes the inventory tractable has two axes: how sensitive the data an app reaches is, and how many people hold edit rights on it. Apps in the sensitive-data, many-editor corner are the first block of work, and one team's spreadsheet with a single editor can sit. Reviewing 2,000 apps at five minutes each takes about 167 hours ^[25].