Agent × IM: Bring execution results into real collaboration
Agent advances a task to a verifiable result; IM brings progress, confirmations, and results into the collaboration path the team already uses.
The question is not only whether Agent can execute
Many tools can generate answers or change code, but the execution process remains in a separate window. Teams still have to restate progress, copy results, and reconfirm ownership, so work breaks between “completed” and “used.”
Put intent, execution, confirmation, and results on one chain
Human intent
→ Desktop interaction
→ Agent work and verification
↔ IM progress and confirmation
→ Result reused by the teamThe point is not to force two modules together, but to exchange explicit state when Agent and IM need to collaborate: whether work is running, what it is waiting for, who must approve, and what the result is.
Two modules, three usage modes
IM ONLYUse messaging and collaboration without starting Agent.
AGENT ONLYHandle tasks and tools without starting IM.
LINKEDExchange messages, progress, approvals, and results through explicit interfaces.
DEVTOOLSOpen observation of events, state, tools, and module communication.
See composition modes and data boundaries in Client architecture.