We built automated testing for vibe-coded apps

2 MatveyF 2 9/18/2025, 7:22:30 PM
Hi HN! We built buffalos.ai because we got tired of users finding bugs we missed.

The problem: AI tools like Cursor made us 10x faster at shipping, but manual testing is still slow.

Buffalo spawns browser agents that click everything users would click, in ways you didn't test. They find the bugs before your users do.

How it works:

1. Paste your staging URL 2. Agents systematically test all interaction paths 3. Get a detailed report with scoring for different category

Would love your feedback.

Free during beta: buffalos.ai

Comments (2)

codingdave · 2h ago
Writing a crawler to hit all possible links and interactions is not the tricky part. Actually understanding the expected behavior, which is not always what the code says it should be, is the tricky part. Without someone actually creating specific assertions and acceptance criteria, this seems like a flawed concept, as it might catch some bone-headed "Oh, clicking here breaks it" mistakes, but those are not the bugs that most teams fight.
MatveyF · 45m ago
Yes, you’re right! Our roadmap and tools are focused on helping non-technical users who build apps with platforms like Lovable, v0, etc. The “bug” isn’t just about clicking a link and it not working. For example, we’re also working on things like: 1. Branding analysis and design drift detection (this happens a lot in vibe-coded designs — e.g., one page uses a purple gradient, another uses a completely different style, which breaks consistency). 2. checking how a new user would actually interact with your app, to see if the flow makes sense.

A crawler alone is just a static check. Our north star is trying to fix the last step from vibe coding to the production, like spawning a browser user agent that actually navigates through the app like a first-time user. Another roadmap idea is to build a browser agent that connects with coding AIs (like Cursor or Claude) to provide automated testing right after a feature is implemented. For example, when we use Claude to build something, we constantly have to flip between the terminal and browser to keep design consistent. The thought is: what if Claude’s code had “eyes” to test what it just built in real time?

Overall, thanks so much for your response. honestly I didn’t expect anyone to reply. This is my first time on HN, and it’s been an awesome interaction!