Agents, in plain English
Not 779 chatbots, and not 779 separate artificial intelligences. Here is what the roster really is and what it can and cannot do.
Plain English
The Empire's roster lists 779 named agents across 50 departments and 12 realms, and the word agent is doing specific work there. An AI agent is a piece of software given one narrow job, the data and tools to do it, and rules about when to stop and hand off to a person. Each agent is named and assigned to a department so its output can be traced back to a single owner rather than to a general-purpose black box. The agents do the continuous numerical work; people set the questions, hold the release gate, and carry the blame.
New to any of this? Read the plain-English glossary — model, agent, hallucination, backtest, edge, closing line and the rest, defined in one place.
Questions people actually ask
An Empire agent is software with one narrow duty inside one department — pricing a market, watching a line, checking a claim, drafting a briefing — plus the data it needs and a rule about when to stop. Each agent has a name, a realm, a specialty and a status, and every card shows its signal duty, its collaborators and its coordination network.
No. They are 779 configured roles, not 779 independently invented AI systems. Think of one capable engine with 779 job descriptions attached: each role has its own instructions, its own data, its own limits and its own name on the output. That structure exists so a wrong answer can be traced to one duty and fixed, instead of disappearing into a single general chatbot.
It works without being asked. A chatbot waits for a question and forgets afterwards; an Empire agent holds a standing duty, runs on the desk's schedule, keeps the context of its department, and produces output that carries a timestamp and a status into the Member Wire. It can also be graded later, because what it published is on the record.
People do. Agents produce and assemble the work, but a department release passes through review before it reaches members, and the HumanAssist layer exists specifically to add context, correction, review and escalation. Some lanes are marked paper-only and never enter the graded record at all. No agent can move money, place a wager or grant access.
Yes, and the Empire assumes it. A hallucination is fluent, confident, invented output — a statistic that does not exist, a player who did not play — produced because the underlying model generates plausible text rather than looking facts up. That is why outputs are timestamped, graded against real outcomes, and put in front of human reviewers instead of trusted on sight.
Accountability and legibility. A named agent in a named department means every published line has an owner, a scope and a history you can follow, and it means members can tell which desk produced what. It also keeps the system honest with itself: when a duty performs badly, there is a specific thing to fix rather than a vague complaint about the AI.
AI agents are software and can be wrong. Every record-eligible output is graded against real outcomes, and nothing produced by an agent is a guarantee. Sports content is 21+ where legal — 1-800-GAMBLER. Market and retirement content is not financial, investment, tax or legal advice.