Meetup: Coding Agents & Pizza! ๐๐ค
@alongubkin: Secure AI workers
@assafbenjosef: Fixing CI & spinning up live VMs
@nir_naim: Real eval suites over generic benchmarks
RSVP ๐ https://t.co/FjTLoWIIke
happy to announce RewardHackBench, built on @harborframework , we study if sandboxes can stop agents from cheating on benchmarks
https://t.co/YT4Ibq7tE9
thinking about hosting dinner in NY for all sandbox builders
I know we all compete but it's like the early days of cloud computing and it'd be nice if we met each other in person before we start racing
who should I invite?
@ivanburazin@davidcrawshaw@tereza_tizkova@HeyGarrison@AniC_dev@mlejva
created landscape of agent benchmarks, and ranked them by maturity.
https://t.co/Kz8rg5RYEa
Opus 4.8 release made it clearer to me that people don't always know about the different benchmarks.
open to PRs
run Claude Managed Agents on https://t.co/b2wt3lluWX.
anthropic runs the harness, islo runs the isolated runtimes, gateway egress, secrets outside the agent, persistence across idle/resume.
https://t.co/5e7lFlTA8b
You should run Claude Code isolated from your main machine (especially if using --dangerously-skip-permissions). Here are your options: https://t.co/3auIowiciJ