Testing apps with TypeSafe AI jev & OpenCode will be soooo fast small demo:

@typesafeai @opencode what is Jev's role here?

@typesafeai @opencode what does jev do there? is this basically giving jev a list of things to do and jev chooses the next action?

@typesafeai @opencode @injaneity also had some great results using cua driver

@typesafeai @opencode @grok what does Jev do? i have access i’ll test later but from everything youve gathered what is this

@typesafeai @opencode nice. does opencode step back in after every click? i put the click/check loop in a runner for my codex version. curious how you split it

@typesafeai @opencode is Jev doing the computer use here?

@typesafeai @opencode I found a coding-agent session where 99% of 21.8M tokens were repeated context. Would love to throw Jev at the same transcript and see if it can distinguish what’s actually worth carrying forward from what’s just context baggage.

@typesafeai @opencode Fast iteration and AI-assisted testing is a powerful combo.

@typesafeai @opencode GLiNER can do the same while being free, open weights and local

@typesafeai @opencode Just made it play Snake and Tetris!

@typesafeai @opencode This works because it only got options to click, every clickable hyperlink was listed, and named in a way that makes it easy to understand. Jev here didn't interact with the browser in any way. Maybe a good approach would be having AI controlling playwright as a fallback.

@typesafeai @opencode looks like playwright-style a11y tree short ids, are you disabling playwright's forced wait states to force it to go faster?

@typesafeai @opencode why does this video say gpt 5.6 sol tho???

@typesafeai @opencode one sec, getting this thing into VR lmao

@typesafeai @opencode Fast click-throughs still miss the branch where the agent hits Buy because the label looked right. Keep the same loop, point checkout at a mock, and assert the charge never left the sandbox. Speed without that gate is a demo that will look green in CI and wrong in prod.

@typesafeai @opencode jev putting rocket boosters on top of browseruse

@typesafeai @opencode A decision layer that fast is the whole demo. 🔥

@typesafeai @opencode For an app-testing agent, I would separate navigation success from assertion success. A fast agent reaching the page has not proved the feature works. Seed known defects and check whether it fails the test with evidence. My decision-testing checklist:

@typesafeai @opencode Wait — is Jev choosing every click from the page state, or just scoring candidates that OpenCode already proposed? That 1.5s “planning browser call” is nasty either way.

@typesafeai @opencode This is where Jev starts feeling genuinely useful for dev workflows.

@typesafeai @opencode holy shit this is fast

@typesafeai @opencode @tylerxdev

@typesafeai @opencode @nanotchec

@typesafeai @opencode @Teknium in hermes soon?

@typesafeai @opencode what TUI is that?

@typesafeai @opencode Yet another video with no audio forcing me to guess what I am supposed to care about. I'm tired of reading tell your AI to add audio please.

@typesafeai @opencode just turn-off the reasoonings mode, and get the flash speed lol

@typesafeai @opencode I don't buy it, chatgpt can do it this fast, if not faster, with very specific repetitive tasks just like this. have it actually browse the web, instead of doing prebaked tasks

@typesafeai @opencode Wow amazing, I was legally talking about this!

@typesafeai @opencode how did you set this up? thanks!

@typesafeai @opencode Nice use case

@typesafeai @opencode Where we can test it ?

@typesafeai @opencode Is this what panicked the LLM players this weekend? Efficiency seems inevitable. Sinking trillions into 40 year assets is so high risk

@typesafeai @opencode @grok how to try Jev. I only got Hermes agent,opencode,Pi and Oh My Pi

@typesafeai @opencode The mobile equivalent here is here too!