DEMO / RECORDING
Overview & result
Rafal Wilinski built a massively parallel adversarial browser testing suite that tries to break web application releases using Jev for rapid element choices instead of costly vision models.
From the original post
thanks to @typesafeai's Jev we now have massively parallel browser-based adversarial testing suite that tries to break each release.
and it costs pennies.
How Jev fits in the loop
- Spin up high-concurrency headless browser instances against target staging URLs.
- Extract candidate interactive elements at each DOM state change.
- Jev picks the next action in tens of milliseconds without screenshot latency.
- Record uncaught client exceptions, crash logs, and broken UI states.
How to reproduce
- Inspect the tweet video demonstration showing multiple browser workers launching in parallel.
- Map candidate interactable DOM nodes to Jev Choice prompts without taking screenshots.
- Run stress test suites against staging builds to detect state machine deadlock.
Why this build matters
Demonstrates how Jev makes high-volume, automated agentic QA economically and latency-wise practical for everyday continuous delivery.
Limitations
- Author reported performance figures; JevForAgents has not independently stress-tested the parallel suite.
- Does not replace deterministic integration assertions.
- Complex animations require custom wait triggers.