# How a Coworker Runs

When you publish, OneByZero Neo turns everything you configured into a single running agent. Seeing how it works through a request helps you understand how your settings play out in practice.

| Stage | What happens |
| --- | --- |
| **1. Channel input** | A message arrives from a connected channel (for example NeoGPT, Microsoft Teams, or Outlook), or a scheduled proactive routine fires. |
| **2. Agent** | The agent runs on instructions assembled from your [Basics](basics_identity.html) — objective, scope, persona, and tone — together with the [Behaviour](behaviour.html) procedure and boundary rules, using the foundation model you selected. |
| **3. Skills as tools** | Your [Skills](skills.html) are attached to the agent as tools and called as needed, each chosen from its description. |
| **4. Approval (optional)** | If a skill is gated **Ask** — or a business rule applies — a human approval step pauses execution and sends a request to the configured [Governance](governance.html) channel before the action runs. |
| **5. Output** | The agent replies to the user and, if Channel Delivery is configured, can also push output to a channel route. |

As it works, the Coworker draws on its short-term and long-term memory according to the [Memory](memory.html) tab.

After publishing, you can open **View your orchestrating agent** to see these stages laid out as a diagram in the [Agentic Flow editor](../9_agentic_flow_builder/index.html).