💻 Close the Lid: Taking Coding Agents from Laptop to Factory

​Most teams run coding agents on a laptop, one at a time, with a developer approving every step. That model breaks as soon as you want agents working in parallel, overnight, or in response to a production alarm. Close the lid and the work stops. Give the agent your shell and your credentials go with it.

​This session walks the path from coding agents as dev tools today toward the idea of a governed software factory: autonomous loops that run in parallel and around the clock, inside secure guardrails. We'll dive deep into the two controls that let agents operate safely: the cage (sandbox), which sets what an agent can reach, and the leash (harness), which sets what it can do. Then we'll look at where full automation pays off and where a human should stay in the loop.

​Arran Peterson is a Solutions Architect at AWS, working with enterprise customers across Australia on everything in the cloud, from infrastructure to AI. These are his notes from the field: what he's seen work, what hasn't, and the questions still open in a fast-changing landscape. The examples are built on AWS, but the lessons are transferable to whichever host, agent or model you use now or adopt next.