Leaked screenshots show Anthropic building a full-stack app builder directly inside Claude
The interface includes template starters for chatbots, games, photo albums, and landing pages
It runs on Opus 4.6 with a project management layer and prompt-to-app workflow
This puts Anthropic in direct competition with Lovable, Bolt, and every vibe-coding startup built on their API
Anthropic already has the model advantage, the user base, and the infrastructure to ship this fast.
A leaked screenshot from inside Claude's interface just confirmed what many suspected. Anthropic is building a full-stack app builder directly into Claude. The feature, tagged "Let's ship something great," shows a prompt-to-app workflow with template starters, model selection, and project management built in.
The screenshot surfaced on X via @hysteresis_x with a simple caption: "Sneak leak at something coming soon to Claude." It pulled 587 likes and 40 retweets before the broader AI community picked it up and started dissecting what it means.
If this ships, it puts Anthropic in direct competition with Lovable, Bolt.new, and the entire vibe-coding market. And unlike those startups, Anthropic owns the model they'd build it on.
What the Leaked Screenshot Shows
The interface is clean and intentional. Not a prototype mockup. Not a debug screen someone accidentally left visible. This looks like a product that's close to shipping.
At the top: the Claude logo with a pixel-art character (a viking ship mascot). Below that, the headline "Let's ship something great" in large type. Then a prompt input field with the example text "Build a minimal hello world landing page with a red button." The model selector shows Opus 4.6 with a dropdown to switch models.
Below the prompt field: four template buttons. AI chatbot. Photo album. Space Invaders. Landing page. Each with a distinct icon. These aren't random examples. They represent four different app categories: AI-powered tools, media-heavy apps, games, and marketing pages. That coverage suggests Anthropic is targeting general-purpose app creation, not just one vertical.
At the bottom of the screen: a "Your Projects" section with what appears to be a project list or gallery. This means the builder includes persistence. You create an app, it gets saved, you can return to it later. That's a platform feature, not a one-shot demo. It implies version history, iteration, and the ability to refine apps over multiple sessions instead of generating everything in one prompt.
One telling detail: there's an error toast in the top-right corner reading "Session couldn't be created. You can try again." That's the kind of error you see in production systems under active development. It suggests real infrastructure behind the UI, not just a frontend mockup.
Why Vibe-Coding Startups Should Be Nervous
The vibe-coding market is crowded. Lovable raised money and built a real product. Bolt.new gained traction with developers who want working apps from prompts. v0 by Vercel handles UI generation. Replit has its own deploy-from-prompt flow.
But there's one thing none of them have: their own frontier model.
Lovable runs on Claude. So does a significant chunk of the vibe-coding market. When Anthropic builds a competing product, they're not just entering the market. They're entering it with a structural advantage that nobody else can replicate. They control the model, the inference costs, the context window, and the system prompt. Every competitor using Claude's API is now building on top of their competitor's infrastructure.
This is the platform play that's been predictable since Claude started generating code artifacts in conversations. Artifacts let Claude write and preview code inside the chat window. Then came Claude's "AI-powered apps" feature that let users build and share interactive apps directly in the chat. The full app builder is the next logical step: take that generated code, deploy it to a real URL, give it persistence, and let users manage it as an ongoing project.
Each step has moved Claude from "chatbot that writes code" closer to "platform that builds and hosts software." This leak shows they're almost there.
Anthropic also has something the startups lack: data. Hundreds of millions of conversations showing what people actually try to build with AI. They know the most common prompts, the failure modes, the patterns that work. That's training data for the builder itself, not just the model.
What This Means for Developers
If you're building on top of Claude's API to offer app generation, this is a direct competitive threat. Anthropic shipping a native builder means your product is now competing with your provider. That's not new in tech (AWS did it to half its customer base) but it's the first time it's happened this visibly in the AI coding space.
For solo developers and small teams, this could be a net positive. A Claude-native app builder backed by Opus 4.6 would likely produce higher quality output than third-party tools that add latency, abstraction layers, and their own prompt engineering on top of the same model. Fewer layers between you and the model generally means better results.
I've used Lovable, Bolt, and v0. They all add their own system prompts, guardrails, and deployment wrappers around Claude or GPT. Some of that adds value. Some of it gets in the way. An Anthropic-native builder could skip the middleman entirely and give you raw model capability with purpose-built deployment infrastructure.
The template approach is interesting too. AI chatbot, photo album, Space Invaders, landing page. These four templates cover the categories that make up roughly 80% of what people try to build with vibe-coding tools. Instead of a blank prompt, you get a starting point that the model already understands well. That's a UX decision that reduces the cold-start problem where people stare at an empty prompt and don't know what to ask for.
The model selector showing Opus 4.6 also hints at pricing tiers. You might get basic app generation on Sonnet for free or cheap, with Opus available for complex multi-file applications. That would match Anthropic's existing pricing structure on claude.ai where different models serve different use cases.
What's Still Missing From the Picture
The screenshot shows a UI. It doesn't show deployment. The biggest question is whether Anthropic will host the generated apps or just give you the code. Hosting changes everything. If Claude builds your app and also serves it on a URL, that's a full platform. If it generates code you have to deploy yourself, it's closer to a smarter version of artifacts.
The "Your Projects" section suggests hosting or at least persistent builds. But there's no confirmation of custom domains, databases, authentication, or any of the infrastructure that turns a generated app into a real product.
There's also no timeline. The error toast ("Session couldn't be created") suggests this is still in development. Anthropic hasn't acknowledged the leak publicly. It could ship next week or sit behind a feature flag for months.
The pricing model is also unknown. Lovable charges roughly 18 EUR per month for its starter plan. If Anthropic bundles this into the existing Claude Pro subscription at around 18 EUR per month, the value proposition for standalone vibe-coding tools gets very thin very fast.
There's also the question of code quality. Current vibe-coding tools generate functional apps, but the code itself is often messy, hard to maintain, and difficult to extend manually. If Anthropic's builder produces Claude Code-quality output (which it should, given the shared infrastructure), that alone could be a differentiator. Clean, readable code you can actually hand off to a developer later is something most prompt-to-app tools still struggle with.
Bottom Line
Anthropic building an app builder inside Claude was always a matter of when, not if. The leaked screenshot confirms the "when" is close. The interface is polished. The templates are practical. The infrastructure (projects, model selection, persistence) suggests this is a real product, not an experiment.
For the vibe-coding startups that built their businesses on top of Claude's API, this is the moment they've been preparing for. Some will differentiate on deployment, collaboration, or vertical specialization. Others will find themselves competing directly with their AI provider.
For everyone else, this is good news. More competition in app generation means better tools, lower prices, and faster iteration. Anthropic entering the market with a native builder backed by their own models sets a new baseline for what prompt-to-app should look like.
The app builder market just got its biggest competitor. And it has home-field advantage.
I'll be watching for the official announcement. When this ships, I'm testing it against every vibe-coding tool I've used and writing up the comparison. That should be an interesting one.
Top comments (0)