Defining Agents and Tools
AI team members and tools should be designed together.
If you design the agents first and hand them a vague pile of tools later, or you design the tools first and hope roles will emerge naturally, the workflow gets harder to understand.
Start with the roles
Ask:
- what roles exist in the workspace
- what each role is responsible for
- what each role needs to see and do
Then design the tools that support those roles.
Keep tool access role-specific
A tool being available somewhere in the workspace does not mean every AI team member should use it.
Keep the assignment tight:
- intake roles get intake tools
- review roles get review tools
- release roles get release tools
That makes the workflow easier to govern and easier to debug.
Keep the tool surface structured
Tools should have:
- clear names
- clear input schema
- clear output schema
- clear side-effect boundaries
That is what makes them usable from AI team members, views, hooks, and other packs.