I build the infrastructure that runs untrusted AI agents safely, at scale. Sandboxing, isolation, the boundary between an agent and everything it could reach. Small tools on the weekends, mostly CLIs, always in public.
work
decks
writing
my longest-running side project is a google sheet
tracking every dollar manually in a google sheet since 2020
how we built secure, scalable agent infra on aws
millions of untrusted agents, each in a private micro-vm with zero credentials
everything i got wrong in the last 4,000 commits
every mistake from a year as the first engineer at browser use, in chronological order
a production architecture for the browser use library
save state, re-queue, restore state. surviving the lambda timeout without losing a run
how we built secure, scalable agent sandbox infra
from aws lambda to unikraft micro-vms with a control plane architecture
currently
locationsan francisco, us
readingkvm, vmms, and where the isolation boundary actually sits
in sf? coffee.
always down to meet people building interesting things. agent infra, sandboxing, isolation, open source, startups.
larsen@cundric.com →