OpenHands

Image from iOS.jpg

after-phone-390-menu-inside-viewport.png

before-phone-390-menu-overflows-right-edge.png

I had Opus 5.5 driving Opus 5.5 agents to run OpenHands.

First, it mapped all behaviors it could find existing/intended in the codebase. Wrote them as recipes. It made an easy CLI for agents, to run the recipes; so they don’t need to whip up scripts all the time and fight states. Then they ran it all, found 64 bugs, filed, fixed, reviewed, fixed.

As Graham said, it’s not perfect, but I do think we live in the future…

=&gt; 5 hours initial mapping of OpenHands
=&gt; total took only ~35% weekly Max sub

In an image, an example “recipe” that agent is running via its cli; and an example bug it found and fixed.

Oops, a dialog it discovered :shushing_face:

It was fixed <https://github.com/OpenHands/OpenHands/pull/18040|here>, just some tweak to how the title of 2 plugins is composed

It must be that we didn't test for multiple plugins, though I seem to recall having multiple was intended

TBH I _love_ this. The agent will just go and test things it can do, things that were not yet considered by a human one at a time; this miraculous technology changes the way we thought of the Test Pyramid and all

Really cool to see how much we can do now