Station

Every organisation that deploys AI agents faces the same problem within 90 days: nobody knows who is responsible when something goes wrong. The agent drifts. The output degrades. A customer gets a hallucinated response. And the post-mortem reveals that three people assumed someone else was watching.

A Station is the answer to "who owns this?" before the question gets asked. It is a human seat — not a role title, not a Slack channel, not a RACI matrix entry. It is a defined territory with a single owner, explicit boundaries, and direct accountability for every AI agent operating within it.

Traditional org charts were designed for human-only teams. They describe reporting lines, not operational territory. When AI agents enter the picture, the reporting line is irrelevant — an agent does not report to anyone. It operates within a boundary that someone must own. That someone holds a Station.

A Station has six branches, and all six must be populated for the Station to be real rather than decorative.

Domain Ownership defines what Company Map territory this Station holds. Decision Rights define what falls inside this Station's Mandate — what they can decide without asking anyone, what requires sign-off, and what is explicitly outside their authority. Post Responsibility defines which Posts this Station commissions, owns, and governs. Groundwork Responsibility defines what the Station must complete before any Post under their authority runs.

Rhythm and Cadence defines how this Station operates inside the Waypost cadence — what their Pulse looks like, what they bring to the Bearing. Outcomes defines what this Station is accountable for delivering — a True North Metric, a set of Waypoints, and defined failure conditions.

When every person holds a Station instead of a job title, the accountability gaps disappear. Not the operational gaps — those are always there. The accountability gaps. The situations where something happens and nobody quite owns the response.

In EOS, the closest equivalent is the Accountability Chart seat. The difference is that an EOS seat defines a person's role in a human team. A Station defines a person's territory in a human-agentic team — including governance of AI agents that operate within that territory.

In RACI matrices, a Station owner is always the "A" (Accountable). But RACI does not address what happens when an AI agent operating under that accountability drifts, hallucinates, or exceeds its scope. Stations do — through the Post Roster, the Groundwork protocol, and the Relay mechanism.

In traditional org charts, a Station is closer to a "span of control" than a "reporting line." The question is not "who does this person report to?" but "what territory does this person govern, and what automation operates within it?"

One person can hold multiple Stations, but each Station has exactly one owner. A 10-person company might have 4-6 Stations. A 50-person company might have 12-20. The number is driven by the number of distinct territories that require dedicated human judgment, not by headcount.

The most common mistake is creating Stations that are too broad. A Station called "Technology" that governs 15 AI agents and covers infrastructure, product, and data is not a Station — it is a department label wearing a new name. A well-scoped Station can be described in one sentence and its owner can name every agent operating within it.

The second mistake is shared ownership. Two people co-owning a Station means nobody owns it. When the agent drifts, each person assumes the other is watching. Single ownership is non-negotiable.

The third mistake is creating Stations for agents instead of territories. A Station does not exist because you deployed a chatbot. It exists because "customer intake and qualification" is a territory that requires human judgment, and the chatbot happens to operate within it.

The resistance you will get from your team is predictable: "This feels like a lot of overhead." That reaction is a comparison to the current state, which feels lighter because it involves no documentation. But the current state has a hidden cost: every ambiguous ownership moment costs meeting time, incident response time, and leadership bandwidth that should be spent elsewhere.