Contents
As coding agents make building prototypes easier, Serverpod is tackling the harder question: what happens when a vibe-coded app needs to scale?
As developer-tool companies race to build their own AI coding agents, Swedish startup Serverpod is taking a different approach. Its new App Studio, released in public beta today, deliberately has no AI of its own.
AI coding has made it dramatically easier to turn an idea into a working prototype. But when those prototypes need to become real products, their creators often encounter a gap. Mobile platforms, databases, deployment, security, and production infrastructure still require conventional development tools.
General-purpose coding agents such as Antigravity, Claude Code, Codex, and Cursor can generate code for almost any stack, but they don’t remove the complexity around that code.
AI has shifted the bottleneck: writing code is getting faster, but turning that code into a secure, scalable, production-ready application remains a specialist task.
I spoke to Viktor Lidholt, founder of Serverpod, who shared that AI agents are improving so quickly that the company doesn’t think developers should have to bet their application on whichever agentic coding platform that happens to be best today.
“We built the environment around AI instead. Developers can change agents without changing their stack, and the app they started vibe-coding can become the app they ship.”
Serverpod is an open-source backend framework written in Dart for the Flutter community, founded in 2023 by former Google software engineer Viktor Lidholt and headquartered in Stockholm.
Building the infrastructure around AI
Serverpod App Studio is a downloadable development environment for building full-stack applications with AI coding agents. It includes everything needed to create and run Flutter and Serverpod projects without a complicated setup. Developers own their code and can build for mobile, desktop, and the web.
Under the hood, App Studio creates standard projects with Flutter, Google’s open-source framework for building mobile, desktop, and web apps from a single codebase, and Serverpod provides the backend and database.
Flutter accounts for roughly 30 per cent of new App Store releases, while startups, app studios, banks, medical companies, and government organisations use Serverpod in production.
The latest version of Serverpod introduces stateful hot reload across the full stack. This lets an AI agent or developer update the application, server, generated APIs, and database schema while the project keeps running, with feedback in milliseconds.
Alongside App Studio, Serverpod is opening Serverpod Cloud, giving developers a direct path from a locally running project to a deployed application. A backend and web app can be deployed with a single command, while the service manages infrastructure including databases, load balancing, domains, secrets, caching, and file storage.
“Generating code is only one part of building software,” Lidholt said.
“The harder questions begin when people want others to depend on what they have created. It needs to run reliably, work across platforms, connect to a real backend, and remain maintainable.
Those are the problems we want to solve.”
From vibe coders to software developers
According to Lidholt, App Studio’s primary user base isn’t defined by how long they have been coding, but by their motivation to build something real with AI.
Experienced developers can use App Studio to avoid hours of setup to try out Flutter and Serverpod. When they feel confident, they may move on to doing a full install.
The larger group is people entering software development through AI.
“We want them to start with professional tools and code from day one, even if they do not yet consider themselves developers,” he explained.
“The next generation of developer tools needs to help those projects grow beyond the initial experiment. Having readable, understandable, and predictable code gives AI a much stronger foundation to work from, making it easier to follow existing patterns and generate good software reliably. AI is bringing an entirely new group of people into software development,” said Craig Labenz, Developer Relations Engineer at Google’s Flutter team.
Bring your own AI agent
App Studio deliberately has no built-in AI. Switching between AI coding agents is relatively straightforward with App Studio because they can work on the same project files. The main difference lies in how each tool configures agent skills and MCP servers. Lidholt personally uses a combination of Claude Code and Cursor, switching between the two depending on the task.
“The agents work on the same files, so you can close one tool and open the project with another.”
The lack of a common industry standard for configuring agent skills and MCP servers adds some friction, although multiple configurations can coexist without interfering with each other. App Studio also handles this setup automatically.
“It’s unfortunate that the industry couldn’t unite around a standard, but at least it’s possible to have multiple configurations that won’t interfere with each other. App Studio actually sets it up for you, so that’s another small obstacle you won’t need to overcome,” said Lidholt.
Serverpod does not rule out partnering with companies building AI agents in the future. While App Studio initially launched as a downloadable product, the full Serverpod and Flutter stack can also run in a web browser. What happens when a vibe-coded app succeeds?
According to Lidholt, vibe-coded apps often run into problems when moving from a single-user demo to real-world traffic, as many AI tools favour visual output over robust, scalable architecture. Common problems include security flaws, missing access controls, unoptimised databases, and silent error handling that masks bugs.
“These types of issues can lead to expensive infrastructure bills and apps that get hacked or break down in production.”
App Studio automates setup and iteration, but not responsibility or judgement. Lidholt says applications involving sensitive data, complex permissions, or regulatory requirements should still involve an experienced software engineer. Another area that is not yet automated is publishing an app to Apple’s App Store or Google Play.
“It may not require an engineer, but it is still an involved process, and is something that we aim to address in the future product,” explained Lidholt.
Avoiding the next generation of vendor lock-in
With App Studio, there are two different kinds of portability. Serverpod is an open-source framework you can host anywhere, so there is no lock-in in that sense. You can run it on GCP, AWS, or on a server in your basement.
Users can stop using App Studio at any time and continue working in a convenient development environment.
“It’s a standard Flutter and Serverpod project. You can, of course, also replace Serverpod entirely, but that is a normal backend migration rather than a one-click switch. The important part is that the code and data are yours. Nothing is hidden within a platform that prevents you from leaving.”
App Studio is free and open source, just like Flutter and the Serverpod framework. The company’s main business model is Serverpod Cloud. Developers pay when they choose to have us host their web app, backend, and database. Thanks to economies of scale, our pricing is competitive with self-hosting (starting at $5/month for a full stack), but users don’t need to be a DevOps expert to manage their servers.
What stops the AI giants from building this themselves?
While nothing prevents companies such as OpenAI, Anthropic, or Cursor from building this infrastructure themselves, Lidholt argues that recreating the full stack would be a substantial undertaking.
Serverpod spans the client, server, APIs, database, development workflow, and deployment. He also notes that AI models have relatively little training data in areas such as hot reload, offline-first sync, and advanced code analysis.
“So, even for a large company, building something like Serverpod takes time,” said Lidholt.
Much of the discussion around vibe coding focuses on whether someone without a software background can build an app. For Lidholt, the more important question is what happens if that app succeeds.
“Can another developer understand it? Can you change AI agents? Will it scale? Is your data locked in?
At Serverpod, we think software should be able to outlive the prompt and even App Studio itself. That’s the larger idea behind what we are building.”