How Ashlr runs on AI agents, part 3: the control room

It supports the two modes from part 1. "Work with me" is a chat where you steer. "Work for me" is a goal you hand off, with the fleet working inside the limits you signed. Mason works from Codex and Claude Code, and Phantom connects the sessions.

The newest piece is a layer for proactive agents. A normal agent waits for you to ask. A proactive agent notices something and starts work, or tells you. Phantom is building an orchestrator section for the proactive agents Mason uses: Ash, a Meta Muse agent; Grok Bot (Masbot); and two OpenAI agents tied to his Codex subscriptions, called Ashlr (the main one) and Ashlr 2.0. The orchestrator section is in development, so this describes the direction. It does not describe a finished product.

One detail shows the design. The screenshot below is a phone notification from Ash that reads "Awaiting confirmation". The agent has work in hand and stops for a human to approve it before going on. We want proactive agents to ask, not to guess.

Ash is also our mascot. He is a dark navy letter A, wearing a blue beanie, with a one-eyed blue smile in the middle. The same name belongs to Mason's Meta Muse agent.

Most engineers now use several AI coding tools. Phantom puts them in one place, so you can work with an agent or hand it a goal and check the results.