How Ashlr runs on AI agents, part 1: who does what

Masbot is the primary one. Mason calls it Mas. It is his main point of contact for company work. The Engineering Orchestrator manages our software engineers. One engineer agent looks after Phantom and Phantom Secrets. Another looks after the rest of the ecosystem: Locus, Lexicon, Ashlr MD, Ashlr Stack and ashlr-plugin. Commsbot handles email and LinkedIn, and it is almost fully autonomous. The fourth, the Humanoid Golf Orchestrator, runs four sub-agents of its own.

Why a handful of bots and not a hundred? Because every person has limited attention. Mason needs a short list of contacts who each know their area, report back clearly and ask when something is out of bounds. The orchestrators handle the rest of the org chart underneath them.

Real life is messier than a diagram, and our agents make mistakes. When one does, we try to fix it with a rule and not a reminder.

Another honest boundary: our agents do not hold publishing credentials. When two of our tools were ready to release, the work stopped and asked Mason, because there was no npm token or cargo login on the agents' computer. Product decisions, like whether a feature is a product feature, also stay with Mason. The agents can prepare nearly everything. Some decisions stay human.

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.