AI can propose. The runtime decides.
Agents become more useful when they can act. They become safer and more durable when the application still owns what those actions mean and when they are allowed to execute.
In logiCells, an agent is a participant in an executable conceptual system. It can work with goals, knowledge, tools and semantic Actions, but it does not become the authority for business truth or execution merely because it selected a tool.
Semantic identity survives the integration
An Action can remain the business capability while its operationalization routes through native AST to MCP, native code or another service backend. The integration protocol is therefore replaceable without forcing the business vocabulary to follow it.
Discovery is not authorisation
External MCP capabilities enter through catalog and policy boundaries. Effectful operations can suspend as pending confirmations before any dispatch, and approved calls execute through runtime-owned paths rather than model-owned approval.
Knowledge stays scoped
Personal, team, project and organization knowledge can remain distinct while context is assembled for an agent. Evidence, provenance and scope survive the merge so shared knowledge does not become an unqualified flattening of private memory.
Agent sessions, governed execution, MCP integration, native AST operationalization and scoped knowledge are part of the current agent-ready direction of logiCells. See the Agent-ready overview