Workspai.dev

Why Every Agent Needs the Same Workspace

Multiple agents can have different roles while operating from the same governed model and evidence.

Agents do not need identical prompts. They need compatible views of the same workspace.

A coding agent, reviewer, release agent, and incident assistant have different tasks. Their context should be scoped differently. But project identities, contracts, policies, evidence state, and known unknowns should not change with the agent brand or runtime.

Shared substrate, scoped views

Shared agent substrate with scoped viewsTask-specific views differ in emphasis but retain the same governed workspace identities and evidence.

The coding view may emphasize editable scope and commands. The review view may emphasize impact and policy. The release view may emphasize gates and freshness. All views still refer to the same project and evidence identities.

What goes wrong without it

  • one agent edits a directory another agent considers generated;
  • a reviewer checks an obsolete command;
  • a release agent trusts evidence from a different revision;
  • tool-specific instruction files encode conflicting policies;
  • handoffs lose the assumptions that shaped the previous action.

Agent-specific renderers remain useful. AGENTS.md, IDE rules, skills, and MCP resources can express the same governed information in different forms. The renderer is a consumer boundary, not a new source of truth.

This is the purpose of Agent Grounding: give agents a shared operating substrate while preserving their role-specific capabilities.

On this page