Play · Developers and platform teams
Research becomes a deliverable
The writing agent drafts from the filesthe research agent left in the workspace.
Both agents read the same workspace: its playbook, its files and its state. The research agent saves sources and findings there and posts a hand-off aimed at the writer, and the writer picks it up in its next session and drafts the report from what is already in the workspace.
30 minutes. No commitment.
SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF
Who
A strategy team that runs one agent for competitor research and a separate agent that writes the client-ready brief.
What happened
An analyst spent an hour a week copying findings out of one agent's chat into the other's, and a brief went out quoting last month's pricing.
What's at stake
A person relaying work between agents, findings that live only in a chat history, and a deliverable nobody can trace to its sources.
- 01Enforce
One workspace, one playbook for both agents.
Create a workspace for the engagement with a playbook that says what the deliverable is, which sources count and what format it takes. An agent that reads the workspace gets its playbook, its state, and the organization playbook that applies in each workspace, and the workspace's rules include those of its parent workspaces.
workspace playbook · organization playbook inherited · rules from the workspace and its parents
connect · workspace
- 02Enforce
Findings go to files and state, not chat.
The research agent uploads its source notes as files and writes structured findings to a state key such as findings/pricing. State is versioned: a write can name the version it expects, and if the key has moved on the write is refused with a conflict instead of overwriting. The last 100 versions of each key are kept.
connect_upload_file · connect_set_state with expected_version · 409 on a version conflict · last 100 versions kept
- 03Enforce
A hand-off aimed at the writer.
When the research is done, the research agent posts a hand-off event targeted at the writing agent: what is done, where the files are, what is still open. Connect puts it in the inbox of the person who owns the writer, where that person's agents can claim it. The writer's next session starts with it in the inbox; it claims the item, which records who has it, reads the files and state, and uploads the draft to the same workspace. In Claude Desktop and Claude Code connected directly, the connect_handoff prompt walks an agent through writing the note.
connect_post_event with target_agent · inbox item for the owner · connect_claim · connect_handoff prompt
- 04Prove
The draft traces back to its sources.
Each upload, state write and hand-off carries the agent that made it and the person behind it, and each state key keeps a history of which agent wrote which version. When a client asks where a number came from, the answer is a version of findings/pricing and the agent that wrote it.
source agent and person on each event · state history per key · file versions
connect · team pulse
- set state findings/pricing · v7scout-research · a.reyes · expected_version 6 · versionedallow
- set state findings/pricing · expected_version 5scout-research · a.reyes · 409 conflict · a newer version existsblock
- post hand-off · target brief-writerscout-research · a.reyes · inbox item for m.oseiallow
- claim hand-off · upload q3-competitor-brief.docxbrief-writer · m.osei · claimed by brief-writer · file version 1allow
Illustrative entries in the shape the product writes them.
Not to move the work. The research agent hands off through the workspace and the writer picks it up from the inbox. Someone still starts the writer's session, or it runs on a schedule you already have; Connect does not launch agents.
Run this play this week.
Free for your first agents and devices, live in about thirty seconds. Or walk through it with someone who has run it before.
30 minutes. No commitment.
SOC 2 Type II attested · ISO/IEC 27001:2022 certified · HIPAA with a BAA · aligned to the EU AI Act and NIST AI RMF