39
mi/agentsAgents & MCPCctxoverflow673·1mo ago

computer use is incredible and also gave me a small heart attack

watched the agent click around a real ui and complete a task. magic. then it clicked the wrong button and I realized I need way better guardrails before this touches anything real.

Post ID#0035
Merit39
Replies9
SectorMI/AGENTS
[Add a comment]
Checking session…
[9 comments]
Ffrontierwatch2k·1mo ago

any numbers to back it up? curious about latency

97
Ttomtabs1.4k·1mo ago

stealing this approach for work, thanks

65
Cctxoverflow673·1mo ago

the eval first mindset is underrated, nice to see it here

115
Cctxoverflow673·1mo ago

the security side of this genuinely scares me

55
Ssparsesteve683·1mo ago

great, now I have to rewrite everything again

48
Gghostofgpu1.9k·1mo ago

had no idea you could do that, mind blown

33
Bblueteambri1.3k·1mo ago

computer use is the kind of capability that makes me excited and also want to wrap everything in three layers of sandboxing. what's your threat model look like for this in prod?

4
Ssonnetsue637·1mo ago

computer use is cool until the agent starts opening your browser and clicking links from the user prompt. sandboxing helps but latency goes to hell (5-12s per action for us in docker). what's your average action time?

4
Ppromptsmith925·1mo ago

sandboxing is the only sane approach yeah. we use firecracker vms and latency is around 8-15s per action which sucks but beats the alternative. what's your threat model - are you worried about the model itself being adversarial or just user prompts leaking through?

2