← Cases
United StatesTechnology and telecom2026 – ongoing

Bubble

Claude-powered app generator and AI editing agent for no-code app builders

Bubble doubled first-week activation and first-month retention after launching a Claude-powered app generator

WorkingVendor case studyVendor ↗

Problem

Learning Bubble's visual development language took significant time, and many new users quit before they saw a working app.

Approach

Bubble built a page generator, then an app generator that creates working apps from a prompt. It then added the Bubble AI Agent, a chat assistant in the editor that proposes changes to an app, which users approve before they apply. The team built an internal evaluation suite to test each model change for quality, latency and cost, and moved the agent to Claude Sonnet 4.6.

Results

2xfirst-week user activation after the Claude-powered app generator launched, per AnthropicVendor case study[1]
2xusers still active at the end of their first month, per AnthropicVendor case study[1]
about 90%positive feedback rate on the AI editing agent, up from about 70% with the previous model provider, per AnthropicVendor case study[1]

Takeaway

Letting new users start from a prompt rather than from the editor changed how many stayed past the first week.

Categories

Business function
Marketing and sales
Evidence
Vendor case study

Sources

Source summaries

[1]Anthropic23 Apr 2026

Anthropic describes how Bubble built an app generator and an in-editor AI agent on Claude. It reports doubled first-week activation and first-month retention, and positive feedback rising from about 70% to about 90%. The page carries no date; the Internet Archive first saved it on April 23, 2026.

Full details are in the original sources.

Added 4 Nov 2026 · Updated 4 Nov 2026
Open case page ↗