Agents Propose, Humans Publish

AI agents in BFF can participate in your document system — drafting new Policies, SOPs, Work Instructions, or Training, and proposing revisions to documents they author. But the rule is absolute: agents never publish. Every agent write is a proposal that a human approver reviews and publishes.

The Two Controls

Both are set when creating or editing an agent (Team → AI Agents):

  1. Responsible Person (required for every agent) — the human accountable for the agent’s actions and acknowledgements. As the form puts it: “A human is accountable for this agent’s actions and acknowledgements.”
  2. Enable Document Write (off by default) — the opt-in that lets the agent draft documents. The form warns: “Lets this agent draft document changes; a human must approve before publishing.”

You can also give the agent Agent Instructions — standing directives like “Always respond in Spanish” or “Escalate if confidence < 80%”.

How Agent Authoring Works

When a write-enabled agent drafts a document:

  • The agent can author any tier — Policy, SOP, Work Instruction, or Training
  • The agent is recorded as the document’s Author; its responsible human becomes the Owner
  • When the agent updates a draft, the document is automatically submitted for review — it lands in In Review status, waiting for a human
  • The responsible human is notified, with the message prefixed “[On behalf of {agent name}]”

The proposed document appears in the normal documents list with its In Review badge — reviewers and approvers handle it exactly like a human-authored draft, including “Approve & Publish”.

Agents in the Governance Loop

Beyond authoring, agents participate in document control like any team member:

  • They can hold any document role (even Owner) — but publishing always requires a human
  • They can be assigned Required Reading; the responsible human acknowledges on their behalf
  • When a document governing an agent’s task is republished, the agent is notified via webhook so it works from the current version

Where to See What Agents Did

The Agent Activity page (Team group, company admins only) shows execution history across all agents — status, task, steps completed, duration, and a drill-down with each step’s output and decisions. Each agent’s detail panel also has its own Execution History section.

Tip: Treat Enable Document Write like a real authoring privilege. Grant it to agents whose job is documentation (e.g., drafting WIs from observed task runs), and leave it off for pure execution agents — every write-enabled agent adds review load for its responsible human.