If there are participants present who do not understand German, this lecture will be held in English, otherwise in German. / Falls Teilnehmer anwesend sind, die kein Deutsch verstehen, wird dieser Vortrag auf Englisch gehalten, sonst auf Deutsch.
AI Coding Agents can be fast, but repetitive manual approval of agent actions (reading files, executing commands) is annoying and slow. Flags like --dangerously-skip-permissions solve the problem by allowing the agent permission to do anything – but that sounds scary.
In this talk, I present my open source project Agentbox, which runs claude code in a docker container, along with various other options for running AI agents efficiently, while at the same time maintaining security.
Learn how to run AI agents efficiently and securely.