We implemented a new unique feature in our AI harness: initiative for AI assistants

Most AI assistants still wait for the user to ask the next question. In DMJBot, we added a feature we call initiativity: the assistant can periodically wake up, review recent context, and decide whether there is something useful it can prepare on its own. For example, it can continue investigating an unfinished issue, summarize yesterday's work, check project status, or prepare a draft or next-step list before the user asks. The important part is control. It is disabled by default, has a 0-10 initiative level, daily token limits, and custom instructions like "do not send emails", "do not post publicly", or "only prepare summaries." I think this changes the assistant from a passive chat box into something closer to a lightweight AI coworker, while still keeping boundaries around what it can do. I’ll put the blog post link in the comments.