Why vibe coders need tests to prevent regressions
Web Dev Cody demonstrates how to add end-to-end tests with Playwright to a vibe-coded app, showing why testing is essential to prevent AI-assisted coding from silently breaking existing functionality.
Score breakdown
Vibe coders shipping AI-generated code to production can adopt Playwright end-to-end tests — with mocked third-party services — to catch regressions before they reach users, without incurring real API costs on every test run.
- 01Web Dev Cody demonstrates adding Playwright end-to-end tests to a vibe-coded app called 'Video Crafter' using Cursor.
- 02The core problem: larger AI-assisted codebases have a higher chance of breaking existing functionality with each new prompt.
- 03The recommended happy-path test covers login, dashboard navigation, video creation, and viewing the finished video.
Web Dev Cody opens the video by describing a Discord user's problem: their vibe-coded application was increasingly breaking existing features with every new AI prompt, and real users were noticing. The video frames unit testing, integration testing, and end-to-end testing as the primary mitigation strategy for this regression problem, and targets viewers who have no prior testing experience.
The practical demo uses Cursor to write a Playwright end-to-end test for a "Video Crafter" app.
The practical demo uses Cursor to write a Playwright end-to-end test for a "Video Crafter" app. The target happy-path scenario covers a user logging in with a pre-seeded account (assumed to have 5,000 credits), navigating to a video generation page, entering a prompt, keeping default settings, clicking generate, and then verifying that a progress screen appears followed by a video details page with downloadable assets. Web Dev Cody notes that this single flow likely represents 95–98% of all meaningful interactions with the app.
A significant portion of the discussion focuses on the challenge of third-party service integrations — 11 Labs for audio, Fal AI for text-to-video, and OpenAI for script generation — and the need to mock those services during testing to avoid burning real API credits on every test run. For model selection within Cursor, the video suggests using a more capable model for writing end-to-end tests, citing the nuance involved in the task.
Key facts
- 01Web Dev Cody demonstrates adding Playwright end-to-end tests to a vibe-coded app called 'Video Crafter' using Cursor.
- 02The core problem: larger AI-assisted codebases have a higher chance of breaking existing functionality with each new prompt.
- 03The recommended happy-path test covers login, dashboard navigation, video creation, and viewing the finished video.
- 04The test assumes the user account has 5,000 credits pre-seeded to avoid credit-related failures.
- 05The app integrates with 11 Labs (audio), Fal AI (text-to-video), and OpenAI (script generation), all of which need to be mocked in tests.
- 06Mocking third-party services is highlighted as necessary to avoid real API costs on every test run.
- 07The video recommends using a more powerful model in Cursor for writing end-to-end tests due to their complexity.