Skills in. Agents out.

The Agent Factory uses OM2's organizational knowledge graph to observe real workflows across your tools. It identifies patterns: repeatable tasks, manual data movement between systems, and work that follows consistent steps. It doesn't guess from templates — it learns from how your specific teams actually work.

Yes. The Agent Factory proposes agents based on observed workflows. Humans review, approve, and can modify the agent before it deploys. Nothing runs autonomously until a human gives the green light. After deployment, corrections and overrides feed back into the system to improve the agent.

Shared agents handle organizational workflows — the Monday pipeline review, account health reports, incident retrospectives. They serve a team or the whole company. Personalized agents are tailored to individual roles and working patterns — your specific email follow-up style, your report format preferences, your data sources.

Every human correction, override, and approval is training data. The system classifies corrections as data errors (OM2 needs better context) or reasoning errors (the agent needs better instructions). Different feedback channels drive different improvement mechanisms. Agents get measurably better week over week.

Yes. The Agent Factory supports both ambient (auto-discovered) and manual agent creation. You can define custom workflows, skills, and scoring logic. The system will still improve these agents through the same feedback loop.