Shared Conversations
Pick up where a teammate left off. Hand off a task with the full conversation, the work product, and the sources — no re-prompting.
A Shared Conversation is what an AI chat looks like when it isn't trapped in one person's window. The thread, the work product the thread produced, and every source the agent read along the way all live together in the Agent Workspace and can move between teammates without losing context.
This is the third of the three shared assets that make team-level AI work — alongside Skills and the Knowledge Graph.
What a Conversation Holds
Each conversation in the workspace is a self-contained unit of work:
- The thread — Every back-and-forth between teammates and the agent, in order.
- The work product — The draft brief, the email, the deck, the code change — whichever artifact the conversation is producing.
- The sources — Which Knowledge Graph entries were used, which Connector events arrived, which docs the agent read.
- The Skill that's running — Which versioned Skill this conversation is an instance of (if any).
- The owner — Who's accountable for outputs.
When someone opens the conversation, they see the whole bundle. There's nothing to reconstruct.
Handing Off
Handoff is one click. Reassign owner; the conversation lands in the new owner's Inbox with everything intact. Common patterns:
- End-of-day rollover — A teammate ships partial work, hands the conversation to the timezone that's coming online.
- Escalation — A junior analyst escalates to the senior reviewer with the full draft, the agent's reasoning, and the contested source attached.
- Spec → build — Product hands a finalized brief conversation to engineering. The brief, the discussion, the linked Jira tickets all move together.
- Out of office — On the way out the door, pick the conversations you want covered, hand them off, walk out.
The receiver doesn't ask "what is this and where do I start?" They open it and the answer is already on screen.
Cross-Team, Cross-Function
Conversations don't respect functional boundaries the way tools usually do. A single conversation can pull in:
- Sales context from a CRM Connector
- Product truth from the Knowledge Graph
- A draft from a Marketing-authored Skill
- A review gate from Legal's protection rules
- The owner from the Customer Success team
It's how multi-disciplinary work actually happens, given a surface that supports it.
Privacy and Scope
By default, conversations live in your personal scope. You can share with a specific teammate, a team, or org-wide. Shared scope writes through the Governance Engine — protection rules decide who can see, edit, or hand off.
Personal conversations stay private to you, even from your admin.
Why It Matters
Most enterprise AI work today happens in a chat window nobody else can see. The win for the user is real and the loss for the team is invisible. Shared Conversations make that loss visible and recoverable: the same thread is forkable, joinable, and accountable.
It's also where the Knowledge Graph gets most of its new entries from. A conversation that produced a great launch brief becomes a candidate to promote the brief, the reasoning, and the pattern into the graph — so the next conversation starts higher up.
Related
- Agent Workspace — where conversations live and move
- Shared Skills — what a conversation typically runs
- Knowledge Graph — where conversation outputs get promoted to
- Governance Engine — scope, sharing, and review rules
- Output Ledger — records every accepted, edited, or discarded output