Our portfolio company @arcee_ai released a tool called NAC to help agents work through long engineering tasks. The team built it for work they needed to do themselves, including running experiments and working on their infrastructure. One part of how it works seems pretty relevant to anyone spending time with agents. As an agent works on a task, it can build up a lot of history. Some of that is important to whatever happens next. Plenty of it is just the mess of getting there. NAC lets the immediate working context be temporary while keeping a record of the progress so far. The next part of the work can pick up from that record. If you've ever handed a project to another person, you’ve experienced this same problem first-hand. You know they need enough information to get started, but sending them everything at once can be a lot to manage. I'm curious how this develops as we give agents longer and longer assignments.