Danny Graviet(@dtgrav):We did research on whether full sandbox snapshots actually help coding agents recover from failed long-horizon tasks. > The surprising result was that on Terminal-Bench, retrying from scratch often worked better. > However, snapshots became much more interesting when progress lived outside Git, like databases and running services.

Danny Graviet

@dtgrav

ml research engineer @daytonaio | sandboxes | ml systems | evaluations

参加 August 2017

153 フォロー中 186 ファン

Danny Graviet@dtgrav

2026.09.18 19:05

We did research on whether full sandbox snapshots actually help coding agents recover from failed long-horizon tasks. > The surprising result was that on Terminal-Bench, retrying from scratch often worked better. > However, snapshots became much more interesting when progress lived outside Git, like databases and running services.