The system in one page
Learn the few relationships that make the rest of Delegate predictable.
Delegate becomes easier to use once you see how its main parts fit together. The center of the system is a card: a bounded outcome with context, owners, history, and a place for the result.
Work has a home
A team owns people, agents, shared resources, policies, and billing. Inside it, projects collect related work. A project's board organizes cards into visible work streams and stages, with a backlog for work that is not yet scheduled.
A card is the unit of delegation. It can hold a milestone description, tasks, dependencies, labels, assignees, attachments, linked files, comments, activity, and cost. If work matters enough to track or hand off, give it a card.
Agents have a job and a boundary
An agent is more than a model choice. Its persona explains its durable job and methods. Knowledge provides reference material. Memory retains concise, useful facts. Skills describe repeatable procedures. Tools provide actual capabilities. A working style determines how often the agent pauses for your review.
The agent's authority is always constrained by both its configured ceiling and the permissions of the person delegating. Project scope, network rules, credential policy, and approval gates narrow it further.
Assignment and delegation are separate
Assigning an agent makes it available on the card and signals responsibility. It does not silently start work. Mention the agent in a card comment or activity message when you want it to begin, and include any last instructions there.
That creates a run in the card's work stream. The run records the conversation and important events: tools used, files changed, browser actions, questions, lifecycle changes, and cost. You can guide, stop, or continue the work from that record. Parent and sub-agent work stays connected.
Information has different shapes
- Attachments are copies placed on a card for that piece of work.
- Linked files point to shared team-drive content that can evolve.
- Wikis hold project knowledge with history.
- Data tools support structured records, queries, charts, and dashboards.
- Agent knowledge and memory help an agent find or retain context across work without mixing every fact into its permanent instructions.
Choose the shape based on whether the information is task-specific, shared, structured, or durable.
Automation still uses the same system
Scheduled tasks run recurring prompts in team or agent chat under their creator's authority. Monitors react to a card, column, or workspace condition in the context where they are attached. Email, Slack, team chat, and configured SMS channels create governed conversations for verified senders. An external MCP client such as Codex or Claude Code is different again: it claims one specific card and works as a selected agent. Each route has a visible work record, but only card-scoped routes automatically return to a card.
A useful mental model
Think in three questions:
- Where does this outcome live? Put it in the right team, project, and card.
- Who is equipped and allowed to do it? Choose the person or agent, then check context, capabilities, and boundaries.
- How will we know it is done? State the result and evidence up front, then review the work stream and output against them.