CTO @Timeline_Tech. I also maintain Sedum (https://xitter.cf/t.co/nNm7UJhbwm)
Montevideo, Uruguay
Joined July 2009
- Tweets1.1K
- Following1K
- Followers551
- Likes663
Nacho Gutierrez retweeted
The thing I’m most grateful to Elon for is bringing back an age of wonder and excitement to humanity. That was jaw-dropping awesome! Screams and mouths wide open in the audience. Nuts.
Nacho Gutierrez retweeted
I have asked my clanker to send some PRs for the Elixir one. Please ask your clanker to verify them and tell your clanker we are allowed to deviate from the Rails version in the same way Express/Go/Rust/ do (it was objecting to that in other PRs)!
The clanker says there is more work to be done but I don't have infinite tokens, so it would be nice to see those merged first.
Nice! Jev is great for this.
We went the same way with Sedum (open source, on Playwright), and the probability scores made failing tests much easier to trust.
github.com/sedum-dev/sedum
Just added Clef support to my OSS library. Of all the system-one models I’ve tested, Clef is the first one that feels genuinely on par with Jev.
github.com/sedum-dev/sedum
For those adversarial review loops I like Swamp Club and the concept of blackbox UAT tests
youtu.be/BL561UDdeoA?si=Xdnk…
Could be useful for Omarchy releases :)
100% agree. If review is done manual there is a big mismatch in the development pipeline. Is like connecting Starship water deluge system (ie AI code generation) to the pipes of a house. It will not handle the pressure.
Worth adding that a key characteristic for AI end-to-end tests is to bring the cost way down to increase the run frequency.
That's why I've built sedum.dev (Open Source too-slightly different) on top of Jev.
Nacho Gutierrez retweeted
made a small thing: s1, a TUI for testing System One (jev-like) models.
fill in a state, pick a type, write an instruction, get a typed answer back: noul, choice, or score; with calibrated confidence. all from the terminal.
github.com/tnaftali/s1-tui
For end-to-end browser tests written in plain english we are talking 50x cheaper (in the worst case) and 5x faster.
Jev 🩷
Lo uso para end-to-end tests escritos en plain english.
Escribí esta librería para reemplazar todos los flaky e2e tests en mi trabajo y correr los tests relacionados al cambio en cada push de una manera económica
github.com/sedum-dev/sedum
sedum.dev
I'm CTO at @Timeline_Tech. We're shipping more than ever, all of it written with agents, and we needed e2e tests that keep up: easy to grow, resilient, and cheap enough to run all the time.
So I built Sedum on top of Jev. Thanks @typesafeai for making the numbers work.
It's alpha, so the format will change. Works on macOS, Linux and Windows. MIT. Open Source. Bring-your-own key.
npx sedum init
github.com/sedum-dev/sedum
sedum.dev
Jev only reads text. Four identical "Buy" buttons, and you mean the one on the blue tote? It won't guess.
With --vision, a vision model looks at a screenshot and picks one of the real buttons.
A failing test writes a report.md your coding agent can act on: sentence, file:line, scores, screenshot, rerun command.
Claude Code fixing one, sped up 4x.
Upcoming feature: a probability that a failure is a real bug and not a flaky test.
Every check gets a score from Jev (by @typesafeai), not a yes or no.
So a weak pass gets flagged instead of going silently green, a flaky check shows its score dropping before it breaks the build, and one threshold decides how strict you are.